diff --git a/src/main/cpp/server.hpp b/src/main/cpp/server.hpp
index d3c19f06..1e9604f0 100644
--- a/src/main/cpp/server.hpp
+++ b/src/main/cpp/server.hpp
@@ -3624,17 +3624,26 @@ struct server_context {
     }
 
     json model_meta() const {
+        // Read optional string metadata from GGUF headers; empty string if absent.
+        auto read_meta_str = [&](const char * key) -> std::string {
+            char buf[512] = {};
+            int32_t n = llama_model_meta_val_str(model, key, buf, sizeof(buf));
+            return n >= 0 ? std::string(buf, n) : std::string();
+        };
+
         return json{
-            {"vocab_type",  llama_vocab_type(vocab)},
-            {"n_vocab",     llama_vocab_n_tokens(vocab)},
-            {"n_ctx_train", llama_model_n_ctx_train(model)},
-            {"n_embd",      llama_model_n_embd(model)},
-            {"n_params",    llama_model_n_params(model)},
-            {"size",        llama_model_size(model)},
-            {"modalities",  json{
+            {"vocab_type",   llama_vocab_type(vocab)},
+            {"n_vocab",      llama_vocab_n_tokens(vocab)},
+            {"n_ctx_train",  llama_model_n_ctx_train(model)},
+            {"n_embd",       llama_model_n_embd(model)},
+            {"n_params",     llama_model_n_params(model)},
+            {"size",         llama_model_size(model)},
+            {"modalities",   json{
                 {"vision", mctx ? mtmd_support_vision(mctx) : false},
                 {"audio",  mctx ? mtmd_support_audio(mctx)  : false},
             }},
+            {"architecture", read_meta_str("general.architecture")},
+            {"name",         read_meta_str("general.name")},
         };
     }
 };
diff --git a/src/main/java/de/kherud/llama/InferenceParameters.java b/src/main/java/de/kherud/llama/InferenceParameters.java
index 9d97b91a..70e94401 100644
--- a/src/main/java/de/kherud/llama/InferenceParameters.java
+++ b/src/main/java/de/kherud/llama/InferenceParameters.java
@@ -368,16 +368,7 @@ public InferenceParameters setPenaltyPrompt(String penaltyPrompt) {
 	 */
 	public InferenceParameters setPenaltyPrompt(int[] tokens) {
 		if (tokens.length > 0) {
-			StringBuilder builder = new StringBuilder();
-			builder.append("[");
-			for (int i = 0; i < tokens.length; i++) {
-				builder.append(tokens[i]);
-				if (i < tokens.length - 1) {
-					builder.append(", ");
-				}
-			}
-			builder.append("]");
-			parameters.put(PARAM_PENALTY_PROMPT, builder.toString());
+			parameters.put(PARAM_PENALTY_PROMPT, serializer.buildIntArray(tokens).toString());
 		}
 		return this;
 	}
@@ -408,7 +399,7 @@ public InferenceParameters setIgnoreEos(boolean ignoreEos) {
 	 */
 	public InferenceParameters setTokenIdBias(Map<Integer, Float> logitBias) {
 		if (!logitBias.isEmpty()) {
-			parameters.put(PARAM_LOGIT_BIAS, buildBiasPairArray(logitBias, String::valueOf));
+			parameters.put(PARAM_LOGIT_BIAS, serializer.buildTokenIdBiasArray(logitBias).toString());
 		}
 		return this;
 	}
@@ -428,7 +419,7 @@ public InferenceParameters setTokenIdBias(Map<Integer, Float> logitBias) {
 	 */
 	public InferenceParameters disableTokenIds(Collection<Integer> tokenIds) {
 		if (!tokenIds.isEmpty()) {
-			parameters.put(PARAM_LOGIT_BIAS, buildDisablePairArray(tokenIds, String::valueOf));
+			parameters.put(PARAM_LOGIT_BIAS, serializer.buildDisableTokenIdArray(tokenIds).toString());
 		}
 		return this;
 	}
@@ -448,7 +439,7 @@ public InferenceParameters disableTokenIds(Collection<Integer> tokenIds) {
 	 */
 	public InferenceParameters setTokenBias(Map<String, Float> logitBias) {
 		if (!logitBias.isEmpty()) {
-			parameters.put(PARAM_LOGIT_BIAS, buildBiasPairArray(logitBias, this::toJsonString));
+			parameters.put(PARAM_LOGIT_BIAS, serializer.buildTokenStringBiasArray(logitBias).toString());
 		}
 		return this;
 	}
@@ -468,7 +459,7 @@ public InferenceParameters setTokenBias(Map<String, Float> logitBias) {
 	 */
 	public InferenceParameters disableTokens(Collection<String> tokens) {
 		if (!tokens.isEmpty()) {
-			parameters.put(PARAM_LOGIT_BIAS, buildDisablePairArray(tokens, this::toJsonString));
+			parameters.put(PARAM_LOGIT_BIAS, serializer.buildDisableTokenStringArray(tokens).toString());
 		}
 		return this;
 	}
@@ -481,16 +472,7 @@ public InferenceParameters disableTokens(Collection<String> tokens) {
 	 */
 	public InferenceParameters setStopStrings(String... stopStrings) {
 		if (stopStrings.length > 0) {
-			StringBuilder builder = new StringBuilder();
-			builder.append("[");
-			for (int i = 0; i < stopStrings.length; i++) {
-				builder.append(toJsonString(stopStrings[i]));
-				if (i < stopStrings.length - 1) {
-					builder.append(", ");
-				}
-			}
-			builder.append("]");
-			parameters.put(PARAM_STOP, builder.toString());
+			parameters.put(PARAM_STOP, serializer.buildStopStrings(stopStrings).toString());
 		}
 		return this;
 	}
@@ -503,29 +485,7 @@ public InferenceParameters setStopStrings(String... stopStrings) {
 	 */
 	public InferenceParameters setSamplers(Sampler... samplers) {
 		if (samplers.length > 0) {
-			StringBuilder builder = new StringBuilder();
-			builder.append("[");
-			for (int i = 0; i < samplers.length; i++) {
-				switch (samplers[i]) {
-					case TOP_K:
-						builder.append("\"top_k\"");
-						break;
-					case TOP_P:
-						builder.append("\"top_p\"");
-						break;
-					case MIN_P:
-						builder.append("\"min_p\"");
-						break;
-					case TEMPERATURE:
-						builder.append("\"temperature\"");
-						break;
-				}
-				if (i < samplers.length - 1) {
-					builder.append(", ");
-				}
-			}
-			builder.append("]");
-			parameters.put(PARAM_SAMPLERS, builder.toString());
+			parameters.put(PARAM_SAMPLERS, serializer.buildSamplers(samplers).toString());
 		}
 		return this;
 	}
@@ -567,7 +527,7 @@ public InferenceParameters setChatTemplate(String chatTemplate) {
 	 * @return this builder
 	 */
 	public InferenceParameters setChatTemplateKwargs(java.util.Map<String, String> kwargs) {
-		parameters.put(PARAM_CHAT_TEMPLATE_KWARGS, mapToJsonObject(kwargs));
+		parameters.put(PARAM_CHAT_TEMPLATE_KWARGS, serializer.buildRawValueObject(kwargs).toString());
 		return this;
 	}
 
@@ -581,44 +541,7 @@ public InferenceParameters setChatTemplateKwargs(java.util.Map<String, String> k
      * @return this builder
      */
     public InferenceParameters setMessages(String systemMessage, List<Pair<String, String>> messages) {
-		StringBuilder messagesBuilder = new StringBuilder();
-		messagesBuilder.append("[");
-
-        // Add system message (if provided)
-        if (systemMessage != null && !systemMessage.isEmpty()) {
-			messagesBuilder.append("{\"role\": \"system\", \"content\": ")
-					.append(toJsonString(systemMessage))
-					.append("}");
-			if (!messages.isEmpty()) {
-				messagesBuilder.append(", ");
-			}
-        }
-
-        // Add user/assistant messages
-        for (int i = 0; i < messages.size(); i++) {
-            Pair<String, String> message = messages.get(i);
-            String role = message.getKey();
-            String content = message.getValue();
-
-            if (!role.equals("user") && !role.equals("assistant")) {
-                throw new IllegalArgumentException("Invalid role: " + role + ". Role must be 'user' or 'assistant'.");
-            }
-
-			messagesBuilder.append("{\"role\":")
-					.append(toJsonString(role))
-					.append(", \"content\": ")
-					.append(toJsonString(content))
-					.append("}");
-
-			if (i < messages.size() - 1) {
-				messagesBuilder.append(", ");
-			}
-        }
-
-		messagesBuilder.append("]");
-
-        // Convert ArrayNode to a JSON string and store it in parameters
-        parameters.put(PARAM_MESSAGES, messagesBuilder.toString());
+        parameters.put(PARAM_MESSAGES, serializer.buildMessages(systemMessage, messages).toString());
         return this;
     }
 
@@ -627,38 +550,5 @@ InferenceParameters setStream(boolean stream) {
 		return this;
 	}
 
-	private static <K, V> String buildBiasPairArray(Map<K, V> map,
-			java.util.function.Function<K, String> keySerializer) {
-		StringBuilder builder = new StringBuilder("[");
-		int i = 0;
-		for (Map.Entry<K, V> entry : map.entrySet()) {
-			builder.append("[")
-					.append(keySerializer.apply(entry.getKey()))
-					.append(", ")
-					.append(entry.getValue())
-					.append("]");
-			if (i++ < map.size() - 1) {
-				builder.append(", ");
-			}
-		}
-		builder.append("]");
-		return builder.toString();
-	}
-
-	private static <T> String buildDisablePairArray(Collection<T> items,
-			java.util.function.Function<T, String> serializer) {
-		StringBuilder builder = new StringBuilder("[");
-		int i = 0;
-		for (T item : items) {
-			builder.append("[")
-					.append(serializer.apply(item))
-					.append(", false]");
-			if (i++ < items.size() - 1) {
-				builder.append(", ");
-			}
-		}
-		builder.append("]");
-		return builder.toString();
-	}
-
 }
+
diff --git a/src/main/java/de/kherud/llama/JsonParameters.java b/src/main/java/de/kherud/llama/JsonParameters.java
index fc3e9dd2..aaa87297 100644
--- a/src/main/java/de/kherud/llama/JsonParameters.java
+++ b/src/main/java/de/kherud/llama/JsonParameters.java
@@ -1,5 +1,7 @@
 package de.kherud.llama;
 
+import de.kherud.llama.json.ParameterJsonSerializer;
+
 import java.util.HashMap;
 import java.util.Map;
 
@@ -14,6 +16,8 @@ abstract class JsonParameters {
 	// The JNI code for a proper Java-typed data object is comparatively too complex and hard to maintain.
 	final Map<String, String> parameters = new HashMap<>();
 
+	protected final ParameterJsonSerializer serializer = new ParameterJsonSerializer();
+
 	@Override
 	public String toString() {
 		StringBuilder builder = new StringBuilder();
@@ -35,73 +39,8 @@ public String toString() {
 		return builder.toString();
 	}
 
-	static String mapToJsonObject(Map<String, String> map) {
-		StringBuilder sb = new StringBuilder("{");
-		boolean first = true;
-		for (Map.Entry<String, String> entry : map.entrySet()) {
-			if (!first) sb.append(",");
-			sb.append("\"").append(entry.getKey()).append("\":").append(entry.getValue());
-			first = false;
-		}
-		sb.append("}");
-		return sb.toString();
-	}
-
-	// taken from org.json.JSONObject#quote(String, Writer)
 	String toJsonString(String text) {
 		if (text == null) return null;
-		StringBuilder builder = new StringBuilder((text.length()) + 2);
-
-		char b;
-		char c = 0;
-		String hhhh;
-		int i;
-		int len = text.length();
-
-		builder.append('"');
-		for (i = 0; i < len; i += 1) {
-			b = c;
-			c = text.charAt(i);
-			switch (c) {
-				case '\\':
-				case '"':
-					builder.append('\\');
-					builder.append(c);
-					break;
-				case '/':
-					if (b == '<') {
-						builder.append('\\');
-					}
-					builder.append(c);
-					break;
-				case '\b':
-					builder.append("\\b");
-					break;
-				case '\t':
-					builder.append("\\t");
-					break;
-				case '\n':
-					builder.append("\\n");
-					break;
-				case '\f':
-					builder.append("\\f");
-					break;
-				case '\r':
-					builder.append("\\r");
-					break;
-				default:
-					if (c < ' ' || (c >= '\u0080' && c < '\u00a0') || (c >= '\u2000' && c < '\u2100')) {
-						builder.append("\\u");
-						hhhh = Integer.toHexString(c);
-						builder.append("0000", 0, 4 - hhhh.length());
-						builder.append(hhhh);
-					}
-					else {
-						builder.append(c);
-					}
-			}
-		}
-		builder.append('"');
-		return builder.toString();
+		return serializer.toJsonString(text);
 	}
 }
diff --git a/src/main/java/de/kherud/llama/LlamaIterable.java b/src/main/java/de/kherud/llama/LlamaIterable.java
index 7e6dff89..24384db4 100644
--- a/src/main/java/de/kherud/llama/LlamaIterable.java
+++ b/src/main/java/de/kherud/llama/LlamaIterable.java
@@ -3,13 +3,44 @@
 import org.jetbrains.annotations.NotNull;
 
 /**
- * An iterable used by {@link LlamaModel#generate(InferenceParameters)} that specifically returns a {@link LlamaIterator}.
+ * An {@link Iterable} wrapper around {@link LlamaIterator} returned by
+ * {@link LlamaModel#generate(InferenceParameters)} and {@link LlamaModel#generateChat(InferenceParameters)}.
+ *
+ * <p>Implements {@link AutoCloseable} so that a try-with-resources block automatically cancels
+ * any in-progress generation when the loop exits early (e.g. via {@code break}), preventing the
+ * native task slot from leaking:
+ *
+ * <pre>{@code
+ * try (LlamaIterable it = model.generate(params)) {
+ *     for (LlamaOutput o : it) {
+ *         if (done) break;   // close() cancels the native task automatically
+ *     }
+ * }
+ * }</pre>
+ *
+ * <p>A plain for-each loop without try-with-resources continues to work; the {@link #close()}
+ * method just will not be called on early exit in that case.
  */
-@FunctionalInterface
-public interface LlamaIterable extends Iterable<LlamaOutput> {
+public final class LlamaIterable implements Iterable<LlamaOutput>, AutoCloseable {
+
+    private final LlamaIterator iterator;
+
+    LlamaIterable(LlamaIterator iterator) {
+        this.iterator = iterator;
+    }
 
     @NotNull
     @Override
-    LlamaIterator iterator();
+    public LlamaIterator iterator() {
+        return iterator;
+    }
 
+    /**
+     * Cancels any in-progress generation. Delegates to {@link LlamaIterator#close()}.
+     * Safe to call multiple times.
+     */
+    @Override
+    public void close() {
+        iterator.close();
+    }
 }
diff --git a/src/main/java/de/kherud/llama/LlamaIterator.java b/src/main/java/de/kherud/llama/LlamaIterator.java
index c238298e..b431a54b 100644
--- a/src/main/java/de/kherud/llama/LlamaIterator.java
+++ b/src/main/java/de/kherud/llama/LlamaIterator.java
@@ -1,5 +1,6 @@
 package de.kherud.llama;
 
+import de.kherud.llama.json.CompletionResponseParser;
 import java.util.Iterator;
 import java.util.NoSuchElementException;
 
@@ -7,11 +8,16 @@
  * This iterator is used by {@link LlamaModel#generate(InferenceParameters)} and
  * {@link LlamaModel#generateChat(InferenceParameters)}. In addition to implementing {@link Iterator},
  * it allows to cancel ongoing inference (see {@link #cancel()}).
+ *
+ * <p>{@link LlamaIterator} implements {@link AutoCloseable}. When used via {@link LlamaIterable}
+ * inside a try-with-resources block, {@link #close()} is called automatically on early exit
+ * (e.g. {@code break}), preventing the native task slot from leaking.
  */
-public final class LlamaIterator implements Iterator<LlamaOutput> {
+public final class LlamaIterator implements Iterator<LlamaOutput>, AutoCloseable {
 
     private final LlamaModel model;
     private final int taskId;
+    private final CompletionResponseParser completionParser = new CompletionResponseParser();
 
     private boolean hasNext = true;
 
@@ -38,7 +44,7 @@ public LlamaOutput next() {
             throw new NoSuchElementException();
         }
         String json = model.receiveCompletionJson(taskId);
-        LlamaOutput output = LlamaOutput.fromJson(json);
+        LlamaOutput output = completionParser.parse(json);
         hasNext = !output.stop;
         if (output.stop) {
         	model.releaseTask(taskId);
@@ -53,4 +59,18 @@ public void cancel() {
         model.cancelCompletion(taskId);
         hasNext = false;
     }
+
+    /**
+     * Cancels any in-progress generation if the iterator has not yet reached a stop token.
+     * Safe to call multiple times — subsequent calls are no-ops.
+     *
+     * <p>Prefer using the enclosing {@link LlamaIterable} in a try-with-resources block rather
+     * than calling this directly.
+     */
+    @Override
+    public void close() {
+        if (hasNext) {
+            cancel();
+        }
+    }
 }
diff --git a/src/main/java/de/kherud/llama/LlamaModel.java b/src/main/java/de/kherud/llama/LlamaModel.java
index c633aa48..31b8a763 100644
--- a/src/main/java/de/kherud/llama/LlamaModel.java
+++ b/src/main/java/de/kherud/llama/LlamaModel.java
@@ -1,6 +1,9 @@
 package de.kherud.llama;
 
 import de.kherud.llama.args.LogFormat;
+import de.kherud.llama.json.ChatResponseParser;
+import de.kherud.llama.json.CompletionResponseParser;
+import de.kherud.llama.json.RerankResponseParser;
 import java.lang.annotation.Native;
 import java.nio.charset.StandardCharsets;
 import java.util.HashMap;
@@ -30,6 +33,10 @@ public class LlamaModel implements AutoCloseable {
 	@Native
 	private long ctx;
 
+	private final CompletionResponseParser completionParser = new CompletionResponseParser();
+	private final ChatResponseParser chatParser = new ChatResponseParser();
+	private final RerankResponseParser rerankParser = new RerankResponseParser();
+
 	/**
 	 * Load with the given {@link ModelParameters}. Make sure to either set
 	 * <ul>
@@ -56,7 +63,7 @@ public String complete(InferenceParameters parameters) {
 		parameters.setStream(false);
 		int taskId = requestCompletion(parameters.toString());
 		String json = receiveCompletionJson(taskId);
-		return LlamaOutput.getContentFromJson(json);
+		return completionParser.parse(json).text;
 	}
 
 	/**
@@ -67,7 +74,7 @@ public String complete(InferenceParameters parameters) {
 	 * @return iterable LLM outputs
 	 */
 	public LlamaIterable generate(InferenceParameters parameters) {
-		return () -> new LlamaIterator(this, parameters);
+		return new LlamaIterable(new LlamaIterator(this, parameters));
 	}
 	
 	
@@ -157,7 +164,7 @@ public static String jsonSchemaToGrammar(String schema) {
 	 */
 	public List<Pair<String, Float>> rerank(boolean reRank, String query, String... documents) {
 		String json = handleRerank(query, documents);
-		List<Pair<String, Float>> rankedDocuments = LlamaOutput.parseRerankResults(json);
+		List<Pair<String, Float>> rankedDocuments = rerankParser.parse(json);
 		if (reRank) {
 			rankedDocuments.sort((a, b) -> Float.compare(b.getValue(), a.getValue()));
 		}
@@ -174,12 +181,12 @@ public List<Pair<String, Float>> rerank(boolean reRank, String query, String...
 	 */
 	public LlamaOutput rerank(String query, String... documents) {
 		String json = handleRerank(query, documents);
-		List<Pair<String, Float>> results = LlamaOutput.parseRerankResults(json);
+		List<Pair<String, Float>> results = rerankParser.parse(json);
 		Map<String, Float> probabilities = new HashMap<>();
 		for (Pair<String, Float> pair : results) {
 			probabilities.put(pair.getKey(), pair.getValue());
 		}
-		return new LlamaOutput(query, probabilities, true);
+		return new LlamaOutput(query, probabilities, true, StopReason.EOS);
 	}
 
 	native String handleRerank(String query, String... documents) throws LlamaException;
@@ -222,6 +229,20 @@ public String chatComplete(InferenceParameters parameters) {
 		return handleChatCompletions(parameters.toString());
 	}
 
+	/**
+	 * Run an OpenAI-compatible chat completion and return only the assistant's text content.
+	 * This is the plain-string equivalent of {@link #chatComplete(InferenceParameters)}, which
+	 * returns the raw OAI JSON. Use this when you want the generated text directly, the same
+	 * way {@link #complete(InferenceParameters)} works for raw completions.
+	 *
+	 * @param parameters the inference parameters including messages
+	 * @return the assistant's reply text (extracted from {@code choices[0].message.content})
+	 * @throws LlamaException if the model was loaded in embedding mode or if inference fails
+	 */
+	public String chatCompleteText(InferenceParameters parameters) {
+		return chatParser.extractChoiceContent(chatComplete(parameters));
+	}
+
 	/**
 	 * Stream an OpenAI-compatible chat completion token by token. The parameters must contain a
 	 * "messages" array in the standard OpenAI chat format. The model's chat template is automatically applied.
@@ -245,7 +266,7 @@ public String chatComplete(InferenceParameters parameters) {
 	 * @throws LlamaException if inference fails
 	 */
 	public LlamaIterable generateChat(InferenceParameters parameters) {
-		return () -> new LlamaIterator(this, parameters, true);
+		return new LlamaIterable(new LlamaIterator(this, parameters, true));
 	}
 
 	/**
diff --git a/src/main/java/de/kherud/llama/LlamaOutput.java b/src/main/java/de/kherud/llama/LlamaOutput.java
index 00793235..328413f9 100644
--- a/src/main/java/de/kherud/llama/LlamaOutput.java
+++ b/src/main/java/de/kherud/llama/LlamaOutput.java
@@ -2,10 +2,6 @@
 
 import org.jetbrains.annotations.NotNull;
 
-import java.util.ArrayList;
-import java.util.Collections;
-import java.util.HashMap;
-import java.util.List;
 import java.util.Map;
 
 /**
@@ -27,153 +23,25 @@ public final class LlamaOutput {
     @NotNull
     public final Map<String, Float> probabilities;
 
-    final boolean stop;
+    /** Whether this is the final token of the generation. */
+    public final boolean stop;
 
-    LlamaOutput(@NotNull String text, @NotNull Map<String, Float> probabilities, boolean stop) {
+    /**
+     * The reason generation stopped. {@link StopReason#NONE} on intermediate streaming tokens.
+     * Only meaningful when {@link #stop} is {@code true}.
+     */
+    @NotNull
+    public final StopReason stopReason;
+
+    public LlamaOutput(@NotNull String text, @NotNull Map<String, Float> probabilities, boolean stop, @NotNull StopReason stopReason) {
         this.text = text;
         this.probabilities = probabilities;
         this.stop = stop;
+        this.stopReason = stopReason;
     }
 
     @Override
     public String toString() {
         return text;
     }
-
-    /**
-     * Parse a LlamaOutput from a JSON string returned by the native receiveCompletionJson method.
-     * The JSON has the structure: {"content": "...", "stop": true/false, ...}
-     */
-    static LlamaOutput fromJson(String json) {
-        String content = getContentFromJson(json);
-        boolean stop = json.contains("\"stop\":true");
-        Map<String, Float> probabilities = parseProbabilities(json);
-        return new LlamaOutput(content, probabilities, stop);
-    }
-
-    /**
-     * Extract the "content" field from a JSON response string.
-     */
-    static String getContentFromJson(String json) {
-        // Find "content":"..." or "content": "..."
-        int keyIdx = json.indexOf("\"content\"");
-        if (keyIdx < 0) {
-            return "";
-        }
-        int colonIdx = json.indexOf(':', keyIdx + 9);
-        if (colonIdx < 0) {
-            return "";
-        }
-        int startQuote = json.indexOf('"', colonIdx + 1);
-        if (startQuote < 0) {
-            return "";
-        }
-        StringBuilder sb = new StringBuilder();
-        for (int i = startQuote + 1; i < json.length(); i++) {
-            char c = json.charAt(i);
-            if (c == '\\' && i + 1 < json.length()) {
-                char next = json.charAt(i + 1);
-                switch (next) {
-                    case '"': sb.append('"'); i++; break;
-                    case '\\': sb.append('\\'); i++; break;
-                    case '/': sb.append('/'); i++; break;
-                    case 'n': sb.append('\n'); i++; break;
-                    case 'r': sb.append('\r'); i++; break;
-                    case 't': sb.append('\t'); i++; break;
-                    case 'b': sb.append('\b'); i++; break;
-                    case 'f': sb.append('\f'); i++; break;
-                    case 'u':
-                        if (i + 5 < json.length()) {
-                            String hex = json.substring(i + 2, i + 6);
-                            sb.append((char) Integer.parseInt(hex, 16));
-                            i += 5;
-                        }
-                        break;
-                    default: sb.append('\\').append(next); i++; break;
-                }
-            } else if (c == '"') {
-                break;
-            } else {
-                sb.append(c);
-            }
-        }
-        return sb.toString();
-    }
-
-    /**
-     * Parse token probabilities from a JSON response. Returns an empty map if no probabilities are present.
-     */
-    private static Map<String, Float> parseProbabilities(String json) {
-        if (!json.contains("\"completion_probabilities\"")) {
-            return Collections.emptyMap();
-        }
-        // For now, return empty map. Full probability parsing can be added later if needed.
-        // The probabilities data is available in the raw JSON for advanced users.
-        return Collections.emptyMap();
-    }
-
-    /**
-     * Parse rerank results from a JSON array string.
-     * Expected format: [{"document": "...", "index": 0, "score": 0.95}, ...]
-     */
-    static List<Pair<String, Float>> parseRerankResults(String json) {
-        List<Pair<String, Float>> results = new ArrayList<>();
-        // Simple parser for the known JSON array structure
-        int idx = 0;
-        while ((idx = json.indexOf("\"document\"", idx)) >= 0) {
-            // Extract document string
-            int colonIdx = json.indexOf(':', idx + 10);
-            int startQuote = json.indexOf('"', colonIdx + 1);
-            int endQuote = findEndQuote(json, startQuote + 1);
-            String document = unescapeJson(json.substring(startQuote + 1, endQuote));
-
-            // Extract score
-            int scoreIdx = json.indexOf("\"score\"", endQuote);
-            if (scoreIdx < 0) break;
-            int scoreColon = json.indexOf(':', scoreIdx + 7);
-            int scoreStart = scoreColon + 1;
-            // Skip whitespace
-            while (scoreStart < json.length() && json.charAt(scoreStart) == ' ') scoreStart++;
-            int scoreEnd = scoreStart;
-            while (scoreEnd < json.length() && (Character.isDigit(json.charAt(scoreEnd)) || json.charAt(scoreEnd) == '.' || json.charAt(scoreEnd) == '-' || json.charAt(scoreEnd) == 'e' || json.charAt(scoreEnd) == 'E' || json.charAt(scoreEnd) == '+')) scoreEnd++;
-            float score = Float.parseFloat(json.substring(scoreStart, scoreEnd));
-
-            results.add(new Pair<>(document, score));
-            idx = scoreEnd;
-        }
-        return results;
-    }
-
-    private static int findEndQuote(String s, int from) {
-        for (int i = from; i < s.length(); i++) {
-            if (s.charAt(i) == '\\') {
-                i++; // skip escaped char
-            } else if (s.charAt(i) == '"') {
-                return i;
-            }
-        }
-        return s.length();
-    }
-
-    private static String unescapeJson(String s) {
-        if (!s.contains("\\")) return s;
-        StringBuilder sb = new StringBuilder(s.length());
-        for (int i = 0; i < s.length(); i++) {
-            char c = s.charAt(i);
-            if (c == '\\' && i + 1 < s.length()) {
-                char next = s.charAt(i + 1);
-                switch (next) {
-                    case '"': sb.append('"'); i++; break;
-                    case '\\': sb.append('\\'); i++; break;
-                    case 'n': sb.append('\n'); i++; break;
-                    case 'r': sb.append('\r'); i++; break;
-                    case 't': sb.append('\t'); i++; break;
-                    default: sb.append(c); break;
-                }
-            } else {
-                sb.append(c);
-            }
-        }
-        return sb.toString();
-    }
 }
diff --git a/src/main/java/de/kherud/llama/ModelMeta.java b/src/main/java/de/kherud/llama/ModelMeta.java
index 0e31ae38..9981603d 100644
--- a/src/main/java/de/kherud/llama/ModelMeta.java
+++ b/src/main/java/de/kherud/llama/ModelMeta.java
@@ -60,6 +60,23 @@ public boolean supportsAudio() {
         return node.at("/modalities/audio").asBoolean(false);
     }
 
+    /**
+     * The model architecture string from GGUF {@code general.architecture} metadata
+     * (e.g. {@code "llama"}, {@code "gemma3"}, {@code "mistral"}).
+     * Returns an empty string if the field is absent in the GGUF file.
+     */
+    public String getArchitecture() {
+        return node.path("architecture").asText("");
+    }
+
+    /**
+     * The human-readable model name from GGUF {@code general.name} metadata.
+     * Returns an empty string if the field is absent in the GGUF file.
+     */
+    public String getModelName() {
+        return node.path("name").asText("");
+    }
+
     /**
      * Returns the underlying {@link JsonNode} for direct access to any field,
      * including fields added in future llama.cpp versions.
diff --git a/src/main/java/de/kherud/llama/ModelParameters.java b/src/main/java/de/kherud/llama/ModelParameters.java
index b1659d74..0aabec8c 100644
--- a/src/main/java/de/kherud/llama/ModelParameters.java
+++ b/src/main/java/de/kherud/llama/ModelParameters.java
@@ -1,6 +1,7 @@
 package de.kherud.llama;
 
 import de.kherud.llama.args.*;
+import de.kherud.llama.json.ParameterJsonSerializer;
 
 /***
  * Parameters used for initializing a {@link LlamaModel}.
@@ -8,6 +9,8 @@
 @SuppressWarnings("unused")
 public final class ModelParameters extends CliParameters {
 
+    private final ParameterJsonSerializer serializer = new ParameterJsonSerializer();
+
     private static final String ARG_FIT = "--fit";
     static final String ARG_POOLING = "--pooling";
     public static final String FIT_ON = "on";
@@ -233,8 +236,7 @@ public ModelParameters setKeep(int keep) {
      * @return this builder
      */
     public ModelParameters disableContextShift() {
-        parameters.put("--no-context-shift", null);
-        return this;
+        return setFlag(ModelFlag.NO_CONTEXT_SHIFT);
     }
 
     /**
@@ -243,8 +245,7 @@ public ModelParameters disableContextShift() {
      * @return this builder
      */
     public ModelParameters enableFlashAttn() {
-        parameters.put("--flash-attn", null);
-        return this;
+        return setFlag(ModelFlag.FLASH_ATTN);
     }
 
     /**
@@ -253,8 +254,7 @@ public ModelParameters enableFlashAttn() {
      * @return this builder
      */
     public ModelParameters disablePerf() {
-        parameters.put("--no-perf", null);
-        return this;
+        return setFlag(ModelFlag.NO_PERF);
     }
 
     /**
@@ -263,8 +263,7 @@ public ModelParameters disablePerf() {
      * @return this builder
      */
     public ModelParameters enableEscape() {
-        parameters.put("--escape", null);
-        return this;
+        return setFlag(ModelFlag.ESCAPE);
     }
 
     /**
@@ -273,8 +272,7 @@ public ModelParameters enableEscape() {
      * @return this builder
      */
     public ModelParameters disableEscape() {
-        parameters.put("--no-escape", null);
-        return this;
+        return setFlag(ModelFlag.NO_ESCAPE);
     }
 
     /**
@@ -283,8 +281,7 @@ public ModelParameters disableEscape() {
      * @return this builder
      */
     public ModelParameters enableSpecial() {
-        parameters.put("--special", null);
-        return this;
+        return setFlag(ModelFlag.SPECIAL);
     }
 
     /**
@@ -293,8 +290,7 @@ public ModelParameters enableSpecial() {
      * @return this builder
      */
     public ModelParameters skipWarmup() {
-        parameters.put("--no-warmup", null);
-        return this;
+        return setFlag(ModelFlag.NO_WARMUP);
     }
 
     /**
@@ -304,8 +300,7 @@ public ModelParameters skipWarmup() {
      * @return this builder
      */
     public ModelParameters setSpmInfill() {
-        parameters.put("--spm-infill", null);
-        return this;
+        return setFlag(ModelFlag.SPM_INFILL);
     }
 
     /**
@@ -318,8 +313,7 @@ public ModelParameters setSamplers(Sampler... samplers) {
         if (samplers.length > 0) {
             StringBuilder builder = new StringBuilder();
             for (int i = 0; i < samplers.length; i++) {
-                Sampler sampler = samplers[i];
-                builder.append(sampler.name().toLowerCase());
+                builder.append(samplers[i].getArgValue());
                 if (i < samplers.length - 1) {
                     builder.append(";");
                 }
@@ -346,8 +340,7 @@ public ModelParameters setSeed(long seed) {
      * @return this builder
      */
     public ModelParameters ignoreEos() {
-        parameters.put("--ignore-eos", null);
-        return this;
+        return setFlag(ModelFlag.IGNORE_EOS);
     }
 
     /**
@@ -561,7 +554,7 @@ public ModelParameters setDynatempExponent(float dynatempExponent) {
      * @return this builder
      */
     public ModelParameters setMirostat(MiroStat mirostat) {
-        parameters.put("--mirostat", String.valueOf(mirostat.ordinal()));
+        parameters.put("--mirostat", mirostat.getArgValue());
         return this;
     }
 
@@ -774,8 +767,7 @@ public ModelParameters setGrpAttnW(int grpAttnW) {
      * @return this builder
      */
     public ModelParameters enableDumpKvCache() {
-        parameters.put("--dump-kv-cache", null);
-        return this;
+        return setFlag(ModelFlag.DUMP_KV_CACHE);
     }
 
     /**
@@ -784,8 +776,7 @@ public ModelParameters enableDumpKvCache() {
      * @return this builder
      */
     public ModelParameters disableKvOffload() {
-        parameters.put("--no-kv-offload", null);
-        return this;
+        return setFlag(ModelFlag.NO_KV_OFFLOAD);
     }
 
     /**
@@ -795,7 +786,7 @@ public ModelParameters disableKvOffload() {
      * @return this builder
      */
     public ModelParameters setCacheTypeK(CacheType type) {
-        parameters.put("--cache-type-k", type.name().toLowerCase());
+        parameters.put("--cache-type-k", type.getArgValue());
         return this;
     }
 
@@ -806,7 +797,7 @@ public ModelParameters setCacheTypeK(CacheType type) {
      * @return this builder
      */
     public ModelParameters setCacheTypeV(CacheType type) {
-        parameters.put("--cache-type-v", type.name().toLowerCase());
+        parameters.put("--cache-type-v", type.getArgValue());
         return this;
     }
 
@@ -838,8 +829,7 @@ public ModelParameters setParallel(int nParallel) {
      * @return this builder
      */
     public ModelParameters enableContBatching() {
-        parameters.put("--cont-batching", null);
-        return this;
+        return setFlag(ModelFlag.CONT_BATCHING);
     }
 
     /**
@@ -848,8 +838,7 @@ public ModelParameters enableContBatching() {
      * @return this builder
      */
     public ModelParameters disableContBatching() {
-        parameters.put("--no-cont-batching", null);
-        return this;
+        return setFlag(ModelFlag.NO_CONT_BATCHING);
     }
 
     /**
@@ -858,8 +847,7 @@ public ModelParameters disableContBatching() {
      * @return this builder
      */
     public ModelParameters enableMlock() {
-        parameters.put("--mlock", null);
-        return this;
+        return setFlag(ModelFlag.MLOCK);
     }
 
     /**
@@ -868,8 +856,7 @@ public ModelParameters enableMlock() {
      * @return this builder
      */
     public ModelParameters disableMmap() {
-        parameters.put("--no-mmap", null);
-        return this;
+        return setFlag(ModelFlag.NO_MMAP);
     }
 
     /**
@@ -879,7 +866,7 @@ public ModelParameters disableMmap() {
      * @return this builder
      */
     public ModelParameters setNuma(NumaStrategy numaStrategy) {
-        parameters.put("--numa", numaStrategy.name().toLowerCase());
+        parameters.put("--numa", numaStrategy.getArgValue());
         return this;
     }
 
@@ -912,7 +899,7 @@ public ModelParameters setGpuLayers(int gpuLayers) {
      * @return this builder
      */
     public ModelParameters setSplitMode(GpuSplitMode splitMode) {
-        parameters.put("--split-mode", splitMode.name().toLowerCase());
+        parameters.put("--split-mode", splitMode.getArgValue());
         return this;
     }
 
@@ -944,8 +931,7 @@ public ModelParameters setMainGpu(int mainGpu) {
      * @return this builder
      */
     public ModelParameters enableCheckTensors() {
-        parameters.put("--check-tensors", null);
-        return this;
+        return setFlag(ModelFlag.CHECK_TENSORS);
     }
 
     /**
@@ -1100,8 +1086,7 @@ public ModelParameters setHfToken(String hfToken) {
      * @return this builder
      */
     public ModelParameters enableEmbedding() {
-        parameters.put("--embedding", null);
-        return this;
+        return setFlag(ModelFlag.EMBEDDING);
     }
 
     /**
@@ -1110,8 +1095,7 @@ public ModelParameters enableEmbedding() {
      * @return this builder
      */
     public ModelParameters enableReranking() {
-        parameters.put("--reranking", null);
-        return this;
+        return setFlag(ModelFlag.RERANKING);
     }
 
     /**
@@ -1164,7 +1148,7 @@ public ModelParameters setChatTemplate(String chatTemplate) {
      * @return this builder
      */
     public ModelParameters setChatTemplateKwargs(java.util.Map<String, String> kwargs) {
-        parameters.put("--chat-template-kwargs", JsonParameters.mapToJsonObject(kwargs));
+        parameters.put("--chat-template-kwargs", serializer.buildRawValueObject(kwargs).toString());
         return this;
     }
 
@@ -1185,8 +1169,7 @@ public ModelParameters setSlotPromptSimilarity(float similarity) {
      * @return this builder
      */
     public ModelParameters setLoraInitWithoutApply() {
-        parameters.put("--lora-init-without-apply", null);
-        return this;
+        return setFlag(ModelFlag.LORA_INIT_WITHOUT_APPLY);
     }
 
     /**
@@ -1195,8 +1178,7 @@ public ModelParameters setLoraInitWithoutApply() {
      * @return this builder
      */
     public ModelParameters disableLog() {
-        parameters.put("--log-disable", null);
-        return this;
+        return setFlag(ModelFlag.LOG_DISABLE);
     }
 
     /**
@@ -1216,8 +1198,7 @@ public ModelParameters setLogFile(String logFile) {
      * @return this builder
      */
     public ModelParameters setVerbose() {
-        parameters.put("--verbose", null);
-        return this;
+        return setFlag(ModelFlag.VERBOSE);
     }
 
     /**
@@ -1237,8 +1218,7 @@ public ModelParameters setLogVerbosity(int verbosity) {
      * @return this builder
      */
     public ModelParameters enableLogPrefix() {
-        parameters.put("--log-prefix", null);
-        return this;
+        return setFlag(ModelFlag.LOG_PREFIX);
     }
 
     /**
@@ -1247,8 +1227,7 @@ public ModelParameters enableLogPrefix() {
      * @return this builder
      */
     public ModelParameters enableLogTimestamps() {
-        parameters.put("--log-timestamps", null);
-        return this;
+        return setFlag(ModelFlag.LOG_TIMESTAMPS);
     }
 
     /**
@@ -1334,8 +1313,7 @@ public ModelParameters setModelDraft(String modelDraft) {
      * @return this builder
      */
     public ModelParameters enableJinja() {
-        parameters.put("--jinja", null);
-        return this;
+        return setFlag(ModelFlag.JINJA);
     }
 
     /**
@@ -1346,8 +1324,7 @@ public ModelParameters enableJinja() {
      * @return this builder
      */
     public ModelParameters setVocabOnly() {
-        parameters.put("--vocab-only", null);
-        return this;
+        return setFlag(ModelFlag.VOCAB_ONLY);
     }
 
     /**
@@ -1361,8 +1338,8 @@ public ModelParameters setVocabOnly() {
      * @return this builder
      */
     public ModelParameters setKvUnified(boolean kvUnified) {
-        parameters.put(kvUnified ? "--kv-unified" : "--no-kv-unified", null);
-        parameters.remove(kvUnified ? "--no-kv-unified" : "--kv-unified");
+        setFlag(kvUnified ? ModelFlag.KV_UNIFIED : ModelFlag.NO_KV_UNIFIED);
+        clearFlag(kvUnified ? ModelFlag.NO_KV_UNIFIED : ModelFlag.KV_UNIFIED);
         return this;
     }
 
@@ -1399,8 +1376,32 @@ public ModelParameters setCacheRamMib(int cacheRamMib) {
      * @return this builder
      */
     public ModelParameters setClearIdle(boolean clearIdle) {
-        parameters.put(clearIdle ? "--clear-idle" : "--no-clear-idle", null);
-        parameters.remove(clearIdle ? "--no-clear-idle" : "--clear-idle");
+        setFlag(clearIdle ? ModelFlag.CLEAR_IDLE : ModelFlag.NO_CLEAR_IDLE);
+        clearFlag(clearIdle ? ModelFlag.NO_CLEAR_IDLE : ModelFlag.CLEAR_IDLE);
+        return this;
+    }
+
+    /**
+     * Enable the given flag, adding it to the active parameter set.
+     * Equivalent to calling the specific named method (e.g. {@link #enableFlashAttn()}
+     * for {@link ModelFlag#FLASH_ATTN}).
+     *
+     * @param flag the flag to enable
+     * @return this builder
+     */
+    public ModelParameters setFlag(ModelFlag flag) {
+        parameters.put(flag.getCliFlag(), null);
+        return this;
+    }
+
+    /**
+     * Remove the given flag from the active parameter set.
+     *
+     * @param flag the flag to remove
+     * @return this builder
+     */
+    public ModelParameters clearFlag(ModelFlag flag) {
+        parameters.remove(flag.getCliFlag());
         return this;
     }
 
diff --git a/src/main/java/de/kherud/llama/StopReason.java b/src/main/java/de/kherud/llama/StopReason.java
new file mode 100644
index 00000000..9e595b0e
--- /dev/null
+++ b/src/main/java/de/kherud/llama/StopReason.java
@@ -0,0 +1,62 @@
+package de.kherud.llama;
+
+/**
+ * The reason why token generation stopped for a {@link LlamaOutput}.
+ *
+ * <ul>
+ *   <li>{@link #NONE} — generation has not stopped yet (intermediate streaming token);
+ *       {@link #getStopType()} returns {@code null}.</li>
+ *   <li>{@link #EOS} — the model produced the end-of-sequence token.</li>
+ *   <li>{@link #STOP_STRING} — a caller-specified stop string was matched.</li>
+ *   <li>{@link #MAX_TOKENS} — the token budget ({@code nPredict} or context limit) was exhausted;
+ *       the response was truncated.</li>
+ * </ul>
+ */
+public enum StopReason {
+
+    /** No stop yet; the {@code "stop_type"} field is absent for intermediate tokens. */
+    NONE(null),
+
+    /** End-of-sequence token produced. Server {@code "stop_type"} value: {@code "eos"}. */
+    EOS("eos"),
+
+    /** A caller-supplied stop string was matched. Server {@code "stop_type"} value: {@code "word"}. */
+    STOP_STRING("word"),
+
+    /** Token budget exhausted. Server {@code "stop_type"} value: {@code "limit"}. */
+    MAX_TOKENS("limit");
+
+    private final String stopType;
+
+    StopReason(String stopType) {
+        this.stopType = stopType;
+    }
+
+    /**
+     * Returns the {@code "stop_type"} string used by the native server for this constant,
+     * or {@code null} for {@link #NONE} (intermediate tokens carry no stop-type field).
+     *
+     * @return the stop-type string, or {@code null} for {@link #NONE}
+     */
+    public String getStopType() {
+        return stopType;
+    }
+
+    /**
+     * Map a raw {@code "stop_type"} string from the native server to a {@link StopReason}.
+     * Pass the already-extracted field value, e.g.
+     * {@code node.path("stop_type").asText("")}.
+     *
+     * @param stopType the raw stop-type string, or {@code null} / empty for absent field
+     * @return the corresponding {@link StopReason}, or {@link #NONE} if unrecognised
+     */
+    public static StopReason fromStopType(String stopType) {
+        if (stopType == null) return NONE;
+        switch (stopType) {
+            case "eos":   return EOS;
+            case "word":  return STOP_STRING;
+            case "limit": return MAX_TOKENS;
+            default:      return NONE;
+        }
+    }
+}
diff --git a/src/main/java/de/kherud/llama/args/CacheType.java b/src/main/java/de/kherud/llama/args/CacheType.java
index 8404ed75..4e18b4f7 100644
--- a/src/main/java/de/kherud/llama/args/CacheType.java
+++ b/src/main/java/de/kherud/llama/args/CacheType.java
@@ -1,15 +1,28 @@
 package de.kherud.llama.args;
 
-public enum CacheType {
-
-    F32,
-    F16,
-    BF16,
-    Q8_0,
-    Q4_0,
-    Q4_1,
-    IQ4_NL,
-    Q5_0,
-    Q5_1
+/**
+ * KV cache quantization type for {@code --cache-type-k} and {@code --cache-type-v}.
+ */
+public enum CacheType implements CliArg {
 
+    F32("f32"),
+    F16("f16"),
+    BF16("bf16"),
+    Q8_0("q8_0"),
+    Q4_0("q4_0"),
+    Q4_1("q4_1"),
+    IQ4_NL("iq4_nl"),
+    Q5_0("q5_0"),
+    Q5_1("q5_1");
+
+    private final String argValue;
+
+    CacheType(String argValue) {
+        this.argValue = argValue;
+    }
+
+    @Override
+    public String getArgValue() {
+        return argValue;
+    }
 }
diff --git a/src/main/java/de/kherud/llama/args/CliArg.java b/src/main/java/de/kherud/llama/args/CliArg.java
new file mode 100644
index 00000000..285be24c
--- /dev/null
+++ b/src/main/java/de/kherud/llama/args/CliArg.java
@@ -0,0 +1,19 @@
+package de.kherud.llama.args;
+
+/**
+ * Implemented by every enum in this package that maps to a CLI argument value.
+ *
+ * <p>The contract: {@link #getArgValue()} returns the exact string accepted by the
+ * corresponding llama.cpp CLI argument (e.g. {@code "q8_0"} for {@code --cache-type-k q8_0}).
+ * Callers pass this string directly to {@code parameters.put("--flag", arg.getArgValue())}
+ * without any post-processing.
+ */
+public interface CliArg {
+
+    /**
+     * Returns the CLI argument value string for this constant.
+     *
+     * @return the value string accepted by the corresponding llama.cpp CLI argument
+     */
+    String getArgValue();
+}
diff --git a/src/main/java/de/kherud/llama/args/GpuSplitMode.java b/src/main/java/de/kherud/llama/args/GpuSplitMode.java
index 0c0cd934..8994c417 100644
--- a/src/main/java/de/kherud/llama/args/GpuSplitMode.java
+++ b/src/main/java/de/kherud/llama/args/GpuSplitMode.java
@@ -1,8 +1,22 @@
 package de.kherud.llama.args;
 
-public enum GpuSplitMode {
+/**
+ * GPU tensor split mode for {@code --split-mode}.
+ */
+public enum GpuSplitMode implements CliArg {
 
-	NONE,
-	LAYER,
-	ROW
+    NONE("none"),
+    LAYER("layer"),
+    ROW("row");
+
+    private final String argValue;
+
+    GpuSplitMode(String argValue) {
+        this.argValue = argValue;
+    }
+
+    @Override
+    public String getArgValue() {
+        return argValue;
+    }
 }
diff --git a/src/main/java/de/kherud/llama/args/MiroStat.java b/src/main/java/de/kherud/llama/args/MiroStat.java
index 5268d9bc..262e0427 100644
--- a/src/main/java/de/kherud/llama/args/MiroStat.java
+++ b/src/main/java/de/kherud/llama/args/MiroStat.java
@@ -1,8 +1,25 @@
 package de.kherud.llama.args;
 
-public enum MiroStat {
+/**
+ * Mirostat sampling mode for {@code --mirostat}.
+ *
+ * <p>The arg values ({@code "0"}, {@code "1"}, {@code "2"}) are the integer strings
+ * accepted by the CLI flag, matching llama.cpp's {@code MIROSTAT_*} constants.
+ */
+public enum MiroStat implements CliArg {
 
-	DISABLED,
-	V1,
-	V2
+    DISABLED("0"),
+    V1("1"),
+    V2("2");
+
+    private final String argValue;
+
+    MiroStat(String argValue) {
+        this.argValue = argValue;
+    }
+
+    @Override
+    public String getArgValue() {
+        return argValue;
+    }
 }
diff --git a/src/main/java/de/kherud/llama/args/ModelFlag.java b/src/main/java/de/kherud/llama/args/ModelFlag.java
new file mode 100644
index 00000000..056b9260
--- /dev/null
+++ b/src/main/java/de/kherud/llama/args/ModelFlag.java
@@ -0,0 +1,115 @@
+package de.kherud.llama.args;
+
+/**
+ * Boolean CLI flags for {@link de.kherud.llama.ModelParameters}.
+ *
+ * <p>Each constant maps to a single CLI argument that takes no value — its presence
+ * alone enables the behaviour. Pass to
+ * {@link de.kherud.llama.ModelParameters#setFlag(ModelFlag)} /
+ * {@link de.kherud.llama.ModelParameters#clearFlag(ModelFlag)} for programmatic control,
+ * or use the named convenience methods (e.g. {@link de.kherud.llama.ModelParameters#enableFlashAttn()}).
+ */
+public enum ModelFlag {
+
+    /** Disable context shift on infinite text generation. */
+    NO_CONTEXT_SHIFT("--no-context-shift"),
+
+    /** Enable Flash Attention. */
+    FLASH_ATTN("--flash-attn"),
+
+    /** Disable internal libllama performance timings. */
+    NO_PERF("--no-perf"),
+
+    /** Process escape sequences (e.g. {@code \\n}, {@code \\t}). */
+    ESCAPE("--escape"),
+
+    /** Do not process escape sequences. */
+    NO_ESCAPE("--no-escape"),
+
+    /** Enable special tokens in output. */
+    SPECIAL("--special"),
+
+    /** Skip warming up the model with an empty run. */
+    NO_WARMUP("--no-warmup"),
+
+    /** Use Suffix/Prefix/Middle infill pattern instead of Prefix/Suffix/Middle. */
+    SPM_INFILL("--spm-infill"),
+
+    /** Ignore end-of-stream token and continue generating. */
+    IGNORE_EOS("--ignore-eos"),
+
+    /** Enable verbose printing of the KV cache. */
+    DUMP_KV_CACHE("--dump-kv-cache"),
+
+    /** Disable KV offload. */
+    NO_KV_OFFLOAD("--no-kv-offload"),
+
+    /** Enable continuous (dynamic) batching. */
+    CONT_BATCHING("--cont-batching"),
+
+    /** Disable continuous batching. */
+    NO_CONT_BATCHING("--no-cont-batching"),
+
+    /** Force system to keep model in RAM rather than swapping or compressing. */
+    MLOCK("--mlock"),
+
+    /** Do not memory-map model (slower load but may reduce pageouts if not using mlock). */
+    NO_MMAP("--no-mmap"),
+
+    /** Enable checking model tensor data for invalid values. */
+    CHECK_TENSORS("--check-tensors"),
+
+    /** Enable embedding use case; use only with dedicated embedding models. */
+    EMBEDDING("--embedding"),
+
+    /** Enable reranking endpoint on server. */
+    RERANKING("--reranking"),
+
+    /** Load LoRA adapters without applying them (apply later via POST /lora-adapters). */
+    LORA_INIT_WITHOUT_APPLY("--lora-init-without-apply"),
+
+    /** Disable logging. */
+    LOG_DISABLE("--log-disable"),
+
+    /** Set verbosity level to infinity (log all messages). */
+    VERBOSE("--verbose"),
+
+    /** Enable prefix in log messages. */
+    LOG_PREFIX("--log-prefix"),
+
+    /** Enable timestamps in log messages. */
+    LOG_TIMESTAMPS("--log-timestamps"),
+
+    /** Enable Jinja templating for chat templates. */
+    JINJA("--jinja"),
+
+    /** Only load the vocabulary for tokenization; no weights are loaded. */
+    VOCAB_ONLY("--vocab-only"),
+
+    /** Enable a single unified KV buffer shared across all sequences. */
+    KV_UNIFIED("--kv-unified"),
+
+    /** Disable the unified KV buffer. */
+    NO_KV_UNIFIED("--no-kv-unified"),
+
+    /** Enable saving and clearing idle slots when a new task starts. */
+    CLEAR_IDLE("--clear-idle"),
+
+    /** Disable saving and clearing idle slots. */
+    NO_CLEAR_IDLE("--no-clear-idle");
+
+    private final String cliFlag;
+
+    ModelFlag(String cliFlag) {
+        this.cliFlag = cliFlag;
+    }
+
+    /**
+     * Returns the CLI argument string for this flag (e.g. {@code "--flash-attn"}).
+     *
+     * @return the CLI flag string
+     */
+    public String getCliFlag() {
+        return cliFlag;
+    }
+}
diff --git a/src/main/java/de/kherud/llama/args/NumaStrategy.java b/src/main/java/de/kherud/llama/args/NumaStrategy.java
index fa7a61b0..c0bd9682 100644
--- a/src/main/java/de/kherud/llama/args/NumaStrategy.java
+++ b/src/main/java/de/kherud/llama/args/NumaStrategy.java
@@ -1,8 +1,22 @@
 package de.kherud.llama.args;
 
-public enum NumaStrategy {
+/**
+ * NUMA optimization strategy for {@code --numa}.
+ */
+public enum NumaStrategy implements CliArg {
 
-	DISTRIBUTE,
-	ISOLATE,
-	NUMACTL
+    DISTRIBUTE("distribute"),
+    ISOLATE("isolate"),
+    NUMACTL("numactl");
+
+    private final String argValue;
+
+    NumaStrategy(String argValue) {
+        this.argValue = argValue;
+    }
+
+    @Override
+    public String getArgValue() {
+        return argValue;
+    }
 }
diff --git a/src/main/java/de/kherud/llama/args/PoolingType.java b/src/main/java/de/kherud/llama/args/PoolingType.java
index f582898f..e38d13f6 100644
--- a/src/main/java/de/kherud/llama/args/PoolingType.java
+++ b/src/main/java/de/kherud/llama/args/PoolingType.java
@@ -17,7 +17,7 @@
  * @see <a href="https://github.com/ggerganov/llama.cpp/blob/b8609/include/llama.h">
  *      llama.cpp b8609 – include/llama.h: {@code llama_pooling_type} enum</a>
  */
-public enum PoolingType {
+public enum PoolingType implements CliArg {
 
     /**
      * Use the model's built-in default pooling type.
diff --git a/src/main/java/de/kherud/llama/args/RopeScalingType.java b/src/main/java/de/kherud/llama/args/RopeScalingType.java
index 138d05be..ae6bdc0a 100644
--- a/src/main/java/de/kherud/llama/args/RopeScalingType.java
+++ b/src/main/java/de/kherud/llama/args/RopeScalingType.java
@@ -1,6 +1,6 @@
 package de.kherud.llama.args;
 
-public enum RopeScalingType {
+public enum RopeScalingType implements CliArg {
 
     UNSPECIFIED("unspecified"),
     NONE("none"),
diff --git a/src/main/java/de/kherud/llama/args/Sampler.java b/src/main/java/de/kherud/llama/args/Sampler.java
index 564a2e6f..21f9abcd 100644
--- a/src/main/java/de/kherud/llama/args/Sampler.java
+++ b/src/main/java/de/kherud/llama/args/Sampler.java
@@ -1,15 +1,28 @@
 package de.kherud.llama.args;
 
-public enum Sampler {
-
-    DRY,
-    TOP_K,
-    TOP_P,
-    TYP_P,
-    MIN_P,
-    TEMPERATURE,
-    XTC,
-    INFILL,
-    PENALTIES
+/**
+ * Sampling algorithm for {@code --samplers} (CLI) and the {@code "samplers"} JSON field.
+ */
+public enum Sampler implements CliArg {
 
+    DRY("dry"),
+    TOP_K("top_k"),
+    TOP_P("top_p"),
+    TYP_P("typ_p"),
+    MIN_P("min_p"),
+    TEMPERATURE("temperature"),
+    XTC("xtc"),
+    INFILL("infill"),
+    PENALTIES("penalties");
+
+    private final String argValue;
+
+    Sampler(String argValue) {
+        this.argValue = argValue;
+    }
+
+    @Override
+    public String getArgValue() {
+        return argValue;
+    }
 }
diff --git a/src/main/java/de/kherud/llama/json/ChatResponseParser.java b/src/main/java/de/kherud/llama/json/ChatResponseParser.java
new file mode 100644
index 00000000..ce7ce230
--- /dev/null
+++ b/src/main/java/de/kherud/llama/json/ChatResponseParser.java
@@ -0,0 +1,89 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+
+import java.io.IOException;
+
+/**
+ * Pure JSON transforms for OAI-compatible chat completion responses.
+ *
+ * <p>All methods are stateless and have zero dependency on JNI, native libraries, or llama
+ * model state — they can be tested with JSON string literals alone (see
+ * {@code ChatResponseParserTest}).
+ *
+ * <p>The native server produces an OAI-compatible chat completion JSON:
+ * <pre>{@code
+ * {
+ *   "id": "chatcmpl-...",
+ *   "object": "chat.completion",
+ *   "choices": [
+ *     {
+ *       "index": 0,
+ *       "message": {"role": "assistant", "content": "Hello!"},
+ *       "finish_reason": "stop"
+ *     }
+ *   ],
+ *   "usage": {"prompt_tokens": 12, "completion_tokens": 5, "total_tokens": 17}
+ * }
+ * }</pre>
+ */
+public class ChatResponseParser {
+
+    /** Shared Jackson mapper; thread-safe and reused across all instances. */
+    public static final ObjectMapper OBJECT_MAPPER = new ObjectMapper();
+
+    /**
+     * Extract the assistant's reply text from an OAI chat completion JSON string.
+     * Navigates {@code choices[0].message.content} via Jackson.
+     *
+     * <p>Returns an empty string when: the JSON is malformed, {@code choices} is absent
+     * or empty, or {@code content} is null/absent.
+     *
+     * @param json OAI-compatible chat completion JSON string
+     * @return the assistant content string, or {@code ""} on any failure
+     */
+    public String extractChoiceContent(String json) {
+        try {
+            return extractChoiceContent(OBJECT_MAPPER.readTree(json));
+        } catch (IOException e) {
+            return "";
+        }
+    }
+
+    /**
+     * Extract the assistant's reply text from a pre-parsed OAI chat completion node.
+     * Navigates {@code choices[0].message.content} via Jackson path API.
+     *
+     * @param node pre-parsed OAI chat completion response node
+     * @return the assistant content string, or {@code ""} if absent
+     */
+    public String extractChoiceContent(JsonNode node) {
+        return node.path("choices").path(0).path("message").path("content").asText("");
+    }
+
+    /**
+     * Read a numeric usage field from the {@code "usage"} object in a chat completion node.
+     * Common field names: {@code "prompt_tokens"}, {@code "completion_tokens"},
+     * {@code "total_tokens"}.
+     *
+     * @param node  the parsed chat completion response
+     * @param field the field name within {@code "usage"}
+     * @return the integer value, or {@code 0} if the field or the {@code "usage"} object is absent
+     */
+    public int extractUsageField(JsonNode node, String field) {
+        return node.path("usage").path(field).asInt(0);
+    }
+
+    /**
+     * Count the number of choices returned in the response.
+     * Returns {@code 0} when the {@code "choices"} array is absent or not an array.
+     *
+     * @param node pre-parsed OAI chat completion response node
+     * @return the number of choices, or {@code 0} if absent
+     */
+    public int countChoices(JsonNode node) {
+        JsonNode choices = node.path("choices");
+        return choices.isArray() ? choices.size() : 0;
+    }
+}
diff --git a/src/main/java/de/kherud/llama/json/CompletionResponseParser.java b/src/main/java/de/kherud/llama/json/CompletionResponseParser.java
new file mode 100644
index 00000000..61591b01
--- /dev/null
+++ b/src/main/java/de/kherud/llama/json/CompletionResponseParser.java
@@ -0,0 +1,117 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import de.kherud.llama.LlamaOutput;
+import de.kherud.llama.StopReason;
+
+import java.io.IOException;
+import java.util.Collections;
+import java.util.HashMap;
+import java.util.Map;
+
+/**
+ * Pure JSON transforms for native completion/streaming responses.
+ *
+ * <p>All methods are stateless and have zero dependency on JNI, native libraries, or llama
+ * model state — they can be tested with JSON string literals alone (see
+ * {@code CompletionResponseParserTest}).
+ *
+ * <p>The native server produces one JSON object per streamed token:
+ * <pre>{@code
+ * {
+ *   "content": "Hello",
+ *   "stop": false,
+ *   "stop_type": "none",
+ *   "completion_probabilities": [
+ *     {"token": "Hello", "bytes": [...], "id": 15043, "prob": 0.82,
+ *      "top_probs": [{"token": "Hi", "bytes": [...], "id": 9932, "prob": 0.1}]}
+ *   ]
+ * }
+ * }</pre>
+ *
+ * <p>This is the Java analogue of {@code json_helpers.hpp} in the C++ layer.
+ */
+public class CompletionResponseParser {
+
+    /** Shared Jackson mapper; thread-safe and reused across all instances. */
+    public static final ObjectMapper OBJECT_MAPPER = new ObjectMapper();
+
+    /**
+     * Parse a {@link LlamaOutput} from a raw JSON string returned by the native
+     * {@code receiveCompletionJson} method. Delegates to {@link #parse(JsonNode)} after
+     * a single {@code readTree} call so the string is parsed only once.
+     *
+     * @param json raw JSON string from the native completion response
+     * @return parsed {@link LlamaOutput}; empty output on parse failure
+     */
+    public LlamaOutput parse(String json) {
+        try {
+            return parse(OBJECT_MAPPER.readTree(json));
+        } catch (IOException e) {
+            return new LlamaOutput("", Collections.<String, Float>emptyMap(), false, StopReason.NONE);
+        }
+    }
+
+    /**
+     * Parse a {@link LlamaOutput} from a pre-parsed {@link JsonNode}.
+     * Callers that already hold a parsed node should prefer this overload to avoid re-parsing.
+     *
+     * @param node pre-parsed completion response node
+     * @return parsed {@link LlamaOutput}
+     */
+    public LlamaOutput parse(JsonNode node) {
+        String content = extractContent(node);
+        boolean stop = node.path("stop").asBoolean(false);
+        Map<String, Float> probabilities = parseProbabilities(node);
+        StopReason stopReason = stop ? StopReason.fromStopType(node.path("stop_type").asText("")) : StopReason.NONE;
+        return new LlamaOutput(content, probabilities, stop, stopReason);
+    }
+
+    /**
+     * Extract the {@code "content"} string from a completion response node.
+     * Returns an empty string if the field is absent.
+     *
+     * @param node completion response node
+     * @return the content string, or {@code ""} if absent
+     */
+    public String extractContent(JsonNode node) {
+        return node.path("content").asText("");
+    }
+
+    /**
+     * Parse the {@code completion_probabilities} array into a {@code token → probability} map.
+     *
+     * <p>Each array entry carries the generated token and either a {@code "prob"} value
+     * (post-sampling mode) or {@code "logprob"} (pre-sampling mode). The nested
+     * {@code top_probs}/{@code top_logprobs} arrays are invisible at the outer entry level
+     * and do not interfere with field lookup.
+     *
+     * <p>Returns an empty map when the field is absent or the array is empty.
+     * Requires {@code InferenceParameters#setNProbs(int)} to be configured before inference.
+     *
+     * @param root the top-level completion response node
+     * @return map from token string to probability; empty when no probability data is present
+     */
+    public Map<String, Float> parseProbabilities(JsonNode root) {
+        JsonNode array = root.path("completion_probabilities");
+        if (!array.isArray() || array.size() == 0) {
+            return Collections.emptyMap();
+        }
+        Map<String, Float> result = new HashMap<String, Float>();
+        for (JsonNode entry : array) {
+            String token = entry.path("token").asText("");
+            if (token.isEmpty()) continue;
+
+            // "prob" (post-sampling) or "logprob" (pre-sampling)
+            JsonNode probNode = entry.path("prob");
+            if (probNode.isMissingNode() || probNode.isNull()) {
+                probNode = entry.path("logprob");
+            }
+            if (probNode.isMissingNode() || probNode.isNull()) continue;
+
+            result.put(token, (float) probNode.asDouble(0.0));
+        }
+        return result.isEmpty() ? Collections.<String, Float>emptyMap() : result;
+    }
+}
diff --git a/src/main/java/de/kherud/llama/json/ParameterJsonSerializer.java b/src/main/java/de/kherud/llama/json/ParameterJsonSerializer.java
new file mode 100644
index 00000000..b09d8749
--- /dev/null
+++ b/src/main/java/de/kherud/llama/json/ParameterJsonSerializer.java
@@ -0,0 +1,244 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.core.JsonProcessingException;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import de.kherud.llama.Pair;
+import de.kherud.llama.args.Sampler;
+
+import java.io.IOException;
+import java.util.Collection;
+import java.util.List;
+import java.util.Map;
+
+/**
+ * Pure JSON builders for inference request parameters.
+ *
+ * <p>All methods are stateless and have zero dependency on JNI, native libraries, or llama
+ * model state — they can be tested with plain Java values alone (see
+ * {@code ParameterJsonSerializerTest}).
+ *
+ * <p>Methods return Jackson {@link ArrayNode} or {@link ObjectNode}. Callers that need a JSON
+ * string (e.g. callers in {@code JsonParameters}) call {@code node.toString()}.
+ *
+ * <p>This class replaces hand-rolled {@code StringBuilder} loops and the
+ * {@code org.json}-derived {@code toJsonString()} escaper previously embedded in
+ * {@code JsonParameters}.
+ */
+public class ParameterJsonSerializer {
+
+    /** Shared Jackson mapper; thread-safe and reused across all instances. */
+    public static final ObjectMapper OBJECT_MAPPER = new ObjectMapper();
+
+    // ------------------------------------------------------------------
+    // String escaping
+    // ------------------------------------------------------------------
+
+    /**
+     * Serialize a Java string to a quoted, properly escaped JSON string literal
+     * (e.g. {@code "hello\nworld"} → {@code "\"hello\\nworld\""}).
+     * Returns {@code "null"} for a {@code null} input.
+     *
+     * <p>Replaces the hand-rolled {@code toJsonString()} method in {@code JsonParameters}.
+     *
+     * @param value the Java string to serialize, or {@code null}
+     * @return a JSON string literal, or {@code "null"} if the input is {@code null}
+     */
+    public String toJsonString(String value) {
+        if (value == null) return "null";
+        try {
+            return OBJECT_MAPPER.writeValueAsString(value);
+        } catch (JsonProcessingException e) {
+            return "null";
+        }
+    }
+
+    // ------------------------------------------------------------------
+    // Message array
+    // ------------------------------------------------------------------
+
+    /**
+     * Build an OAI-compatible {@code messages} array node.
+     *
+     * <p>An optional system message is prepended when non-null and non-empty.
+     * Each message in {@code messages} must have role {@code "user"} or {@code "assistant"}.
+     *
+     * @param systemMessage optional system prompt; skipped when {@code null} or empty
+     * @param messages list of user/assistant message pairs (role as key, content as value)
+     * @return a Jackson {@link ArrayNode} of {@code {"role", "content"}} message objects
+     * @throws IllegalArgumentException if any message has an invalid role
+     */
+    public ArrayNode buildMessages(String systemMessage, List<Pair<String, String>> messages) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        if (systemMessage != null && !systemMessage.isEmpty()) {
+            ObjectNode sys = OBJECT_MAPPER.createObjectNode();
+            sys.put("role", "system");
+            sys.put("content", systemMessage);
+            arr.add(sys);
+        }
+        for (Pair<String, String> message : messages) {
+            String role = message.getKey();
+            String content = message.getValue();
+            if (!"user".equals(role) && !"assistant".equals(role)) {
+                throw new IllegalArgumentException(
+                        "Invalid role: " + role + ". Role must be 'user' or 'assistant'.");
+            }
+            ObjectNode msg = OBJECT_MAPPER.createObjectNode();
+            msg.put("role", role);
+            msg.put("content", content);
+            arr.add(msg);
+        }
+        return arr;
+    }
+
+    // ------------------------------------------------------------------
+    // Simple array builders
+    // ------------------------------------------------------------------
+
+    /**
+     * Build a JSON string array from the given stop strings
+     * (e.g. {@code ["<|endoftext|>", "\n"]}).
+     *
+     * @param stops one or more stop strings
+     * @return a Jackson {@link ArrayNode} of stop string values
+     */
+    public ArrayNode buildStopStrings(String... stops) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (String stop : stops) arr.add(stop);
+        return arr;
+    }
+
+    /**
+     * Build a JSON string array from the given sampler sequence
+     * (e.g. {@code ["top_k", "top_p", "temperature"]}).
+     *
+     * @param samplers one or more samplers in the desired order
+     * @return a Jackson {@link ArrayNode} of sampler name strings
+     */
+    public ArrayNode buildSamplers(Sampler... samplers) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (Sampler sampler : samplers) {
+            arr.add(sampler.getArgValue());
+        }
+        return arr;
+    }
+
+    /**
+     * Build a JSON integer array from a primitive {@code int[]}
+     * (used for penalty-prompt token sequences).
+     *
+     * @param values the token IDs to include
+     * @return a Jackson {@link ArrayNode} of integer values
+     */
+    public ArrayNode buildIntArray(int[] values) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (int v : values) arr.add(v);
+        return arr;
+    }
+
+    // ------------------------------------------------------------------
+    // Logit-bias pair arrays — [[key, value], ...]
+    // ------------------------------------------------------------------
+
+    /**
+     * Build a logit-bias array for integer token IDs:
+     * {@code [[15043, 1.0], [50256, -0.5]]}.
+     *
+     * @param biases map from token ID to logit bias value
+     * @return a Jackson {@link ArrayNode} of {@code [tokenId, biasValue]} pairs
+     */
+    public ArrayNode buildTokenIdBiasArray(Map<Integer, Float> biases) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (Map.Entry<Integer, Float> entry : biases.entrySet()) {
+            ArrayNode pair = OBJECT_MAPPER.createArrayNode();
+            pair.add(entry.getKey());
+            pair.add(entry.getValue());
+            arr.add(pair);
+        }
+        return arr;
+    }
+
+    /**
+     * Build a logit-bias array for string tokens:
+     * {@code [["Hello", 1.0], [" world", -0.5]]}.
+     *
+     * @param biases map from token string to logit bias value
+     * @return a Jackson {@link ArrayNode} of {@code ["token", biasValue]} pairs
+     */
+    public ArrayNode buildTokenStringBiasArray(Map<String, Float> biases) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (Map.Entry<String, Float> entry : biases.entrySet()) {
+            ArrayNode pair = OBJECT_MAPPER.createArrayNode();
+            pair.add(entry.getKey());
+            pair.add(entry.getValue());
+            arr.add(pair);
+        }
+        return arr;
+    }
+
+    /**
+     * Build a disable-token array for integer token IDs:
+     * {@code [[15043, false], [50256, false]]}.
+     *
+     * @param ids collection of integer token IDs to disable
+     * @return a Jackson {@link ArrayNode} of {@code [tokenId, false]} pairs
+     */
+    public ArrayNode buildDisableTokenIdArray(Collection<Integer> ids) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (Integer id : ids) {
+            ArrayNode pair = OBJECT_MAPPER.createArrayNode();
+            pair.add(id);
+            pair.add(false);
+            arr.add(pair);
+        }
+        return arr;
+    }
+
+    /**
+     * Build a disable-token array for string tokens:
+     * {@code [["Hello", false], [" world", false]]}.
+     *
+     * @param tokens collection of token strings to disable
+     * @return a Jackson {@link ArrayNode} of {@code ["token", false]} pairs
+     */
+    public ArrayNode buildDisableTokenStringArray(Collection<String> tokens) {
+        ArrayNode arr = OBJECT_MAPPER.createArrayNode();
+        for (String token : tokens) {
+            ArrayNode pair = OBJECT_MAPPER.createArrayNode();
+            pair.add(token);
+            pair.add(false);
+            arr.add(pair);
+        }
+        return arr;
+    }
+
+    // ------------------------------------------------------------------
+    // Object with pre-serialized JSON values
+    // ------------------------------------------------------------------
+
+    /**
+     * Build a JSON object where each map value is a <em>pre-serialized JSON string</em>
+     * (not a plain Java string). For example, a map entry {@code ("enable_thinking", "true")}
+     * produces {@code {"enable_thinking": true}}, not {@code {"enable_thinking": "true"}}.
+     *
+     * <p>Used for {@code chat_template_kwargs} which stores raw JSON values.
+     * If a value cannot be parsed as JSON, it is stored as a JSON string literal.
+     *
+     * @param map map of key to pre-serialized JSON value strings
+     * @return a Jackson {@link ObjectNode} with each value embedded as a parsed JSON node
+     */
+    public ObjectNode buildRawValueObject(Map<String, String> map) {
+        ObjectNode node = OBJECT_MAPPER.createObjectNode();
+        for (Map.Entry<String, String> entry : map.entrySet()) {
+            try {
+                JsonNode val = OBJECT_MAPPER.readTree(entry.getValue());
+                node.set(entry.getKey(), val);
+            } catch (IOException e) {
+                node.put(entry.getKey(), entry.getValue());
+            }
+        }
+        return node;
+    }
+}
diff --git a/src/main/java/de/kherud/llama/json/RerankResponseParser.java b/src/main/java/de/kherud/llama/json/RerankResponseParser.java
new file mode 100644
index 00000000..87fd2e13
--- /dev/null
+++ b/src/main/java/de/kherud/llama/json/RerankResponseParser.java
@@ -0,0 +1,67 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import de.kherud.llama.Pair;
+
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+
+/**
+ * Pure JSON transforms for native rerank responses.
+ *
+ * <p>All methods are stateless and have zero dependency on JNI, native libraries, or llama
+ * model state — they can be tested with JSON string literals alone (see
+ * {@code RerankResponseParserTest}).
+ *
+ * <p>The native server produces a JSON array of reranked results:
+ * <pre>{@code
+ * [
+ *   {"document": "The quick brown fox", "index": 0, "score": 0.92},
+ *   {"document": "Another document",    "index": 1, "score": 0.43}
+ * ]
+ * }</pre>
+ */
+public class RerankResponseParser {
+
+    /** Shared Jackson mapper; thread-safe and reused across all instances. */
+    public static final ObjectMapper OBJECT_MAPPER = new ObjectMapper();
+
+    /**
+     * Parse rerank results from a raw JSON array string. Delegates to {@link #parse(JsonNode)}
+     * after a single {@code readTree} call.
+     *
+     * @param json raw JSON array string from the native rerank response
+     * @return list of document/score pairs; empty list on parse failure or empty array
+     */
+    public List<Pair<String, Float>> parse(String json) {
+        try {
+            return parse(OBJECT_MAPPER.readTree(json));
+        } catch (IOException e) {
+            return Collections.emptyList();
+        }
+    }
+
+    /**
+     * Parse rerank results from a pre-parsed {@link JsonNode} array.
+     * Each element must contain {@code "document"} (string) and {@code "score"} (number).
+     * Returns an empty list when the node is not an array or is empty.
+     *
+     * @param arr pre-parsed {@link JsonNode} array of rerank result objects
+     * @return list of document/score pairs; empty list if the node is not an array or is empty
+     */
+    public List<Pair<String, Float>> parse(JsonNode arr) {
+        if (!arr.isArray() || arr.size() == 0) {
+            return Collections.emptyList();
+        }
+        List<Pair<String, Float>> results = new ArrayList<Pair<String, Float>>();
+        for (JsonNode entry : arr) {
+            String doc = entry.path("document").asText("");
+            float score = (float) entry.path("score").asDouble(0.0);
+            results.add(new Pair<String, Float>(doc, score));
+        }
+        return results;
+    }
+}
diff --git a/src/test/java/de/kherud/llama/ChatAdvancedTest.java b/src/test/java/de/kherud/llama/ChatAdvancedTest.java
index 4b75e9f0..bffc019e 100644
--- a/src/test/java/de/kherud/llama/ChatAdvancedTest.java
+++ b/src/test/java/de/kherud/llama/ChatAdvancedTest.java
@@ -8,6 +8,7 @@
 
 import de.kherud.llama.args.MiroStat;
 import de.kherud.llama.args.Sampler;
+import de.kherud.llama.json.CompletionResponseParser;
 import org.junit.AfterClass;
 import org.junit.Assert;
 import org.junit.Assume;
@@ -42,6 +43,7 @@
 public class ChatAdvancedTest {
 
     private static final int N_PREDICT = 10;
+    private final CompletionResponseParser completionParser = new CompletionResponseParser();
     private static final String SIMPLE_PROMPT = "def hello():";
 
     private static LlamaModel model;
@@ -144,7 +146,7 @@ public void testSetNProbsStreamingJsonHasProbabilities() {
         while (!done) {
             String json = model.receiveCompletionJson(taskId);
             Assert.assertNotNull("receiveCompletionJson must not be null", json);
-            LlamaOutput output = LlamaOutput.fromJson(json);
+            LlamaOutput output = completionParser.parse(json);
             if (json.contains("\"completion_probabilities\"")) {
                 foundProbabilities = true;
             }
@@ -338,7 +340,7 @@ public void testRequestCompletionDirectStreaming() {
         while (!stopped) {
             String json = model.receiveCompletionJson(taskId);
             Assert.assertNotNull("receiveCompletionJson must not return null", json);
-            LlamaOutput output = LlamaOutput.fromJson(json);
+            LlamaOutput output = completionParser.parse(json);
             sb.append(output.text);
             tokens++;
             if (output.stop) {
diff --git a/src/test/java/de/kherud/llama/ChatScenarioTest.java b/src/test/java/de/kherud/llama/ChatScenarioTest.java
index 9e21128b..17d6decd 100644
--- a/src/test/java/de/kherud/llama/ChatScenarioTest.java
+++ b/src/test/java/de/kherud/llama/ChatScenarioTest.java
@@ -7,6 +7,8 @@
 import java.util.List;
 
 import de.kherud.llama.args.PoolingType;
+import de.kherud.llama.json.ChatResponseParser;
+import de.kherud.llama.json.CompletionResponseParser;
 import org.junit.AfterClass;
 import org.junit.Assert;
 import org.junit.Assume;
@@ -41,6 +43,8 @@
 public class ChatScenarioTest {
 
     private static final int N_PREDICT = 10;
+    private final CompletionResponseParser completionParser = new CompletionResponseParser();
+    private final ChatResponseParser chatParser = new ChatResponseParser();
 
     private static LlamaModel model;
 
@@ -100,6 +104,50 @@ public void testChatCompleteResponseJsonStructure() {
                 response.contains("\"assistant\"") || response.contains("assistant"));
     }
 
+    /**
+     * chatCompleteText() must return only the assistant's plain text, not the OAI JSON wrapper.
+     * The result should be non-empty and must NOT contain the JSON key "choices".
+     */
+    @Test
+    public void testChatCompleteTextReturnsPlainString() {
+        List<Pair<String, String>> messages = new ArrayList<>();
+        messages.add(new Pair<>("user", "Say the word OK."));
+
+        InferenceParameters params = new InferenceParameters("")
+                .setMessages(null, messages)
+                .setNPredict(N_PREDICT)
+                .setSeed(42)
+                .setTemperature(0.0f);
+
+        String text = model.chatCompleteText(params);
+
+        Assert.assertNotNull(text);
+        Assert.assertFalse("chatCompleteText must not be empty", text.isEmpty());
+        Assert.assertFalse("chatCompleteText must not contain OAI JSON wrapper", text.contains("\"choices\""));
+    }
+
+    /**
+     * chatCompleteText() must return the same content as extracting choices[0].message.content
+     * from the raw chatComplete() JSON.
+     */
+    @Test
+    public void testChatCompleteTextMatchesChatCompleteContent() {
+        List<Pair<String, String>> messages = new ArrayList<>();
+        messages.add(new Pair<>("user", "What is 2 plus 2?"));
+
+        InferenceParameters params = new InferenceParameters("")
+                .setMessages("You are a helpful assistant.", messages)
+                .setNPredict(N_PREDICT)
+                .setSeed(42)
+                .setTemperature(0.0f);
+
+        String rawJson = model.chatComplete(params);
+        String text = model.chatCompleteText(params);
+
+        String expected = chatParser.extractChoiceContent(rawJson);
+        Assert.assertEquals("chatCompleteText must match choices[0].message.content", expected, text);
+    }
+
     /**
      * handleChatCompletions can be called directly with a raw JSON string.
      * Verify the response contains valid OAI chat completion fields.
@@ -147,7 +195,7 @@ public void testRequestChatCompletionDirectStreaming() {
         while (!stopped) {
             String json = model.receiveCompletionJson(taskId);
             Assert.assertNotNull("receiveCompletionJson must not return null", json);
-            LlamaOutput output = LlamaOutput.fromJson(json);
+            LlamaOutput output = completionParser.parse(json);
             sb.append(output.text);
             tokens++;
             if (output.stop) {
@@ -226,7 +274,7 @@ public void testChatCompleteWithStopString() {
                 .setSeed(42)
                 .setTemperature(0.0f);
         String unJson = model.chatComplete(unconstrained);
-        String unContent = extractChoiceContent(unJson);
+        String unContent = chatParser.extractChoiceContent(unJson);
 
         // Stopped at "3"
         InferenceParameters stopped = new InferenceParameters("")
@@ -236,7 +284,7 @@ public void testChatCompleteWithStopString() {
                 .setTemperature(0.0f)
                 .setStopStrings("4");
         String stJson = model.chatComplete(stopped);
-        String stContent = extractChoiceContent(stJson);
+        String stContent = chatParser.extractChoiceContent(stJson);
 
         Assert.assertNotNull("Stop-string response must not be null", stJson);
         // Content with stop should be shorter (or at most equal)
@@ -312,7 +360,7 @@ public void testChatCompleteMultiTurnThreeTurns() {
                     .setTemperature(0.0f);
 
             String json = model.chatComplete(params);
-            String content = extractChoiceContent(json);
+            String content = chatParser.extractChoiceContent(json);
 
             Assert.assertNotNull("Turn " + turn + ": response must not be null", json);
             Assert.assertFalse("Turn " + turn + ": content must not be empty", content.isEmpty());
@@ -421,8 +469,8 @@ public void testHandleInfillDirect() {
         String prefix = "def greet(name):\n    \"\"\" ";
         String suffix = "\n    return greeting\n";
 
-        String json = "{\"input_prefix\": " + toJsonString(prefix) +
-                ", \"input_suffix\": " + toJsonString(suffix) +
+        String json = "{\"input_prefix\": " + jsonStr(prefix) +
+                ", \"input_suffix\": " + jsonStr(suffix) +
                 ", \"n_predict\": " + N_PREDICT +
                 ", \"seed\": 42, \"temperature\": 0.0}";
 
@@ -493,8 +541,8 @@ public void testHandleTokenizeWithSpecialTokens() {
         Assert.assertNotNull(withoutSpecial);
         Assert.assertTrue("Both responses must contain 'tokens'", withSpecial.contains("\"tokens\""));
 
-        int countWith    = countTokensInJson(withSpecial);
-        int countWithout = countTokensInJson(withoutSpecial);
+        int countWith    = tokenCount(withSpecial);
+        int countWithout = tokenCount(withoutSpecial);
 
         Assert.assertTrue(
                 "addSpecial=true should produce at least as many tokens as addSpecial=false " +
@@ -522,7 +570,7 @@ public void testHandleDetokenizeRoundTrip() {
         Assert.assertTrue("handleDetokenize response must contain 'content'", response.contains("\"content\""));
 
         // Extract the detokenized text (simple search for content field value)
-        String detokenized = LlamaOutput.getContentFromJson(response);
+        String detokenized = completionParser.parse(response).text;
         // The tokenizer typically prepends a space; check the meaningful content
         Assert.assertTrue(
                 "Detokenized text should contain original content (got: '" + detokenized + "')",
@@ -590,7 +638,7 @@ public void testChatCompleteNPredictOne() {
         String response = model.chatComplete(params);
         Assert.assertNotNull(response);
         Assert.assertFalse("nPredict=1 must still return a non-empty response", response.isEmpty());
-        String content = extractChoiceContent(response);
+        String content = chatParser.extractChoiceContent(response);
         // Content should be at most one token long — just verify it doesn't crash
         Assert.assertNotNull("Content must not be null for nPredict=1", content);
     }
@@ -640,52 +688,24 @@ public void testGenerateChatStopFlagOnFinalToken() {
     // Helpers
     // ------------------------------------------------------------------
 
-    /**
-     * Extract the assistant's content string from an OAI-compatible chat
-     * completion JSON response.
-     * Expected structure: {"choices":[{"message":{"role":"assistant","content":"..."}}]}
-     */
-    private static String extractChoiceContent(String json) {
-        // Find choices[0].message.content
-        int choicesIdx = json.indexOf("\"choices\"");
-        if (choicesIdx < 0) {
-            // Fallback: try plain "content" field (non-OAI response shape)
-            return LlamaOutput.getContentFromJson(json);
-        }
-        // Find "content" after "choices"
-        int contentIdx = json.indexOf("\"content\"", choicesIdx);
-        if (contentIdx < 0) {
-            return "";
+    /** Serialize a string to a JSON string literal using Jackson. */
+    private static String jsonStr(String s) {
+        try {
+            return CompletionResponseParser.OBJECT_MAPPER.writeValueAsString(s);
+        } catch (Exception e) {
+            return "null";
         }
-        // Reuse LlamaOutput's JSON extractor on a substring starting at "content"
-        return LlamaOutput.getContentFromJson(json.substring(contentIdx));
     }
 
-    /**
-     * Count the number of comma-separated elements in the JSON array value
-     * of the "tokens" field. This is a best-effort heuristic — it works for
-     * the simple integer-array format returned by handleTokenize.
-     */
-    private static int countTokensInJson(String json) {
-        int tokensIdx = json.indexOf("\"tokens\"");
-        if (tokensIdx < 0) return 0;
-        int openBracket = json.indexOf('[', tokensIdx);
-        int closeBracket = json.indexOf(']', openBracket);
-        if (openBracket < 0 || closeBracket < 0) return 0;
-        String array = json.substring(openBracket + 1, closeBracket).trim();
-        if (array.isEmpty()) return 0;
-        return array.split(",").length;
-    }
-
-    /** Minimal JSON string escaping for test helper strings. */
-    private static String toJsonString(String s) {
-        if (s == null) return "null";
-        return "\"" + s
-                .replace("\\", "\\\\")
-                .replace("\"", "\\\"")
-                .replace("\n", "\\n")
-                .replace("\r", "\\r")
-                .replace("\t", "\\t")
-                + "\"";
+    /** Count elements in the {@code "tokens"} array of a tokenize response. */
+    private static int tokenCount(String json) {
+        try {
+            com.fasterxml.jackson.databind.JsonNode node =
+                    CompletionResponseParser.OBJECT_MAPPER.readTree(json);
+            com.fasterxml.jackson.databind.JsonNode arr = node.path("tokens");
+            return arr.isArray() ? arr.size() : 0;
+        } catch (Exception e) {
+            return 0;
+        }
     }
 }
diff --git a/src/test/java/de/kherud/llama/InferenceParametersTest.java b/src/test/java/de/kherud/llama/InferenceParametersTest.java
index d63acef9..211706a2 100644
--- a/src/test/java/de/kherud/llama/InferenceParametersTest.java
+++ b/src/test/java/de/kherud/llama/InferenceParametersTest.java
@@ -276,7 +276,7 @@ public void testSetStopStringsSingle() {
 	@Test
 	public void testSetStopStringsMultiple() {
 		InferenceParameters params = new InferenceParameters("").setStopStrings("stop1", "stop2");
-		assertEquals("[\"stop1\", \"stop2\"]", params.parameters.get("stop"));
+		assertEquals("[\"stop1\",\"stop2\"]", params.parameters.get("stop"));
 	}
 
 	@Test
@@ -299,7 +299,7 @@ public void testSetSamplersSingle() {
 	@Test
 	public void testSetSamplersMultiple() {
 		InferenceParameters params = new InferenceParameters("").setSamplers(Sampler.TOP_K, Sampler.TOP_P, Sampler.TEMPERATURE);
-		assertEquals("[\"top_k\", \"top_p\", \"temperature\"]", params.parameters.get("samplers"));
+		assertEquals("[\"top_k\",\"top_p\",\"temperature\"]", params.parameters.get("samplers"));
 	}
 
 	@Test
@@ -396,7 +396,7 @@ public void testDisableTokensEmpty() {
 	@Test
 	public void testSetPenaltyPromptTokenIds() {
 		InferenceParameters params = new InferenceParameters("").setPenaltyPrompt(new int[]{1, 2, 3});
-		assertEquals("[1, 2, 3]", params.parameters.get("penalty_prompt"));
+		assertEquals("[1,2,3]", params.parameters.get("penalty_prompt"));
 	}
 
 	@Test
@@ -532,11 +532,12 @@ public void testToJsonStringNull() {
 	}
 
 	@Test
-	public void testToJsonStringEscapesSlashAfterLt() {
-		// '</' should become '<\/'
+	public void testToJsonStringSlashNotEscaped() {
+		// Jackson does not escape '/' — forward slashes are passed through verbatim
 		InferenceParameters params = new InferenceParameters("</script>");
 		String value = params.parameters.get("prompt");
-		assertTrue(value.contains("<\\/"));
+		assertTrue(value.contains("</script>"));
+		assertFalse(value.contains("<\\/"));
 	}
 
 	// -------------------------------------------------------------------------
diff --git a/src/test/java/de/kherud/llama/LlamaModelTest.java b/src/test/java/de/kherud/llama/LlamaModelTest.java
index 0bd34ccd..80adfa34 100644
--- a/src/test/java/de/kherud/llama/LlamaModelTest.java
+++ b/src/test/java/de/kherud/llama/LlamaModelTest.java
@@ -183,6 +183,30 @@ public void testCancelGenerating() {
 		);
 	}
 
+	/**
+	 * LlamaIterable implements AutoCloseable. Breaking out of a for-each loop early inside a
+	 * try-with-resources block must not throw and must not leave the task slot hanging — the
+	 * iterator's close() cancels the native task automatically.
+	 */
+	@Test
+	public void testGenerateAutoCloseOnEarlyBreak() throws Exception {
+		InferenceParameters params = new InferenceParameters(prefix).setNPredict(nPredict);
+
+		int collected = 0;
+		try (LlamaIterable iterable = model.generate(params)) {
+			for (LlamaOutput ignored : iterable) {
+				collected++;
+				break; // exit before stop token
+			}
+		} // close() must cancel without throwing
+
+		Assert.assertTrue("Should have collected at least one token before break", collected >= 1);
+
+		// The model must still be usable after an early-exit close
+		String result = model.complete(new InferenceParameters(prefix).setNPredict(5));
+		Assert.assertNotNull("Model must be functional after autoclosed iterator", result);
+	}
+
 	@Test
 	public void testEmbedding() {
 		float[] embedding = model.embed(prefix);
@@ -946,6 +970,15 @@ public void testGetModelMeta() throws LlamaException {
 		Assert.assertTrue("modalities field must be present", meta.asJson().has("modalities"));
 		Assert.assertTrue("vocab_type field must be present", meta.asJson().has("vocab_type"));
 
+		// Architecture and name from GGUF general.* metadata
+		String architecture = meta.getArchitecture();
+		Assert.assertNotNull("getArchitecture() must not return null", architecture);
+		Assert.assertFalse("CodeLlama GGUF must have general.architecture set", architecture.isEmpty());
+
+		// general.name may or may not be present in the GGUF; just verify the getter does not throw
+		String modelName = meta.getModelName();
+		Assert.assertNotNull("getModelName() must not return null", modelName);
+
 		// Round-trip: toString() must produce valid compact JSON containing all top-level keys
 		String json = meta.toString();
 		Assert.assertNotNull(json);
@@ -958,10 +991,7 @@ public void testGetModelMeta() throws LlamaException {
 		Assert.assertTrue(json.contains("\"modalities\""));
 		Assert.assertTrue(json.contains("\"vision\""));
 		Assert.assertTrue(json.contains("\"audio\""));
-
-		// Fill in the expected value from the failure message and re-run to pin exact output:
-		Assert.assertEquals("{\"vocab_type\":1,\"n_vocab\":32016,\"n_ctx_train\":16384,"
-				+ "\"n_embd\":4096,\"n_params\":6738546688,\"size\":2825274880,"
-				+ "\"modalities\":{\"vision\":false,\"audio\":false}}", json);
+		Assert.assertTrue(json.contains("\"architecture\""));
+		Assert.assertTrue(json.contains("\"name\""));
 	}
 }
diff --git a/src/test/java/de/kherud/llama/LlamaOutputTest.java b/src/test/java/de/kherud/llama/LlamaOutputTest.java
index e5fa8529..4a4bf730 100644
--- a/src/test/java/de/kherud/llama/LlamaOutputTest.java
+++ b/src/test/java/de/kherud/llama/LlamaOutputTest.java
@@ -1,5 +1,6 @@
 package de.kherud.llama;
 
+import de.kherud.llama.json.CompletionResponseParser;
 import java.util.Collections;
 import java.util.HashMap;
 import java.util.Map;
@@ -9,27 +10,29 @@
 import static org.junit.Assert.*;
 
 @ClaudeGenerated(
-        purpose = "Verify that LlamaOutput correctly stores text, the probability map and stop flag " +
-                  "unchanged, and that toString() delegates to the text field."
+        purpose = "Verify that LlamaOutput correctly stores text, the probability map, stop flag, " +
+                  "and stopReason, and that toString() delegates to the text field."
 )
 public class LlamaOutputTest {
 
+	private final CompletionResponseParser parser = new CompletionResponseParser();
+
 	@Test
 	public void testTextFromString() {
-		LlamaOutput output = new LlamaOutput("hello", Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput("hello", Collections.emptyMap(), false, StopReason.NONE);
 		assertEquals("hello", output.text);
 	}
 
 	@Test
 	public void testEmptyText() {
-		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false, StopReason.NONE);
 		assertEquals("", output.text);
 	}
 
 	@Test
 	public void testUtf8MultibyteText() {
 		String original = "héllo wörld";
-		LlamaOutput output = new LlamaOutput(original, Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput(original, Collections.emptyMap(), false, StopReason.NONE);
 		assertEquals(original, output.text);
 	}
 
@@ -38,7 +41,7 @@ public void testProbabilitiesStored() {
 		Map<String, Float> probs = new HashMap<>();
 		probs.put("hello", 0.9f);
 		probs.put("world", 0.1f);
-		LlamaOutput output = new LlamaOutput("", probs, false);
+		LlamaOutput output = new LlamaOutput("", probs, false, StopReason.NONE);
 		assertEquals(2, output.probabilities.size());
 		assertEquals(0.9f, output.probabilities.get("hello"), 0.0001f);
 		assertEquals(0.1f, output.probabilities.get("world"), 0.0001f);
@@ -46,38 +49,38 @@ public void testProbabilitiesStored() {
 
 	@Test
 	public void testEmptyProbabilities() {
-		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false, StopReason.NONE);
 		assertTrue(output.probabilities.isEmpty());
 	}
 
 	@Test
 	public void testStopFlagFalse() {
-		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false, StopReason.NONE);
 		assertFalse(output.stop);
 	}
 
 	@Test
 	public void testStopFlagTrue() {
-		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), true);
+		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), true, StopReason.EOS);
 		assertTrue(output.stop);
 	}
 
 	@Test
 	public void testToStringReturnsText() {
-		LlamaOutput output = new LlamaOutput("generated text", Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput("generated text", Collections.emptyMap(), false, StopReason.NONE);
 		assertEquals("generated text", output.toString());
 	}
 
 	@Test
 	public void testToStringEmptyText() {
-		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false);
+		LlamaOutput output = new LlamaOutput("", Collections.emptyMap(), false, StopReason.NONE);
 		assertEquals("", output.toString());
 	}
 
 	@Test
 	public void testFromJson() {
 		String json = "{\"content\":\"hello world\",\"stop\":true}";
-		LlamaOutput output = LlamaOutput.fromJson(json);
+		LlamaOutput output = parser.parse(json);
 		assertEquals("hello world", output.text);
 		assertTrue(output.stop);
 	}
@@ -85,14 +88,122 @@ public void testFromJson() {
 	@Test
 	public void testFromJsonWithEscapes() {
 		String json = "{\"content\":\"line1\\nline2\\t\\\"quoted\\\"\",\"stop\":false}";
-		LlamaOutput output = LlamaOutput.fromJson(json);
+		LlamaOutput output = parser.parse(json);
 		assertEquals("line1\nline2\t\"quoted\"", output.text);
 		assertFalse(output.stop);
 	}
 
+	@Test
+	public void testFromJsonWithUnicodeEscape() {
+		String json = "{\"content\":\"caf\\u00e9\",\"stop\":false}";
+		LlamaOutput output = parser.parse(json);
+		assertEquals("café", output.text);
+		assertFalse(output.stop);
+	}
+
+	@Test
+	public void testFromJsonMalformedReturnsEmptyNonStop() {
+		LlamaOutput output = parser.parse("{not valid json");
+		assertEquals("", output.text);
+		assertFalse(output.stop);
+		assertEquals(StopReason.NONE, output.stopReason);
+		assertTrue(output.probabilities.isEmpty());
+	}
+
 	@Test
 	public void testGetContentFromJsonEmpty() {
 		String json = "{\"content\":\"\",\"stop\":true}";
-		assertEquals("", LlamaOutput.getContentFromJson(json));
+		assertEquals("", parser.parse(json).text);
+	}
+
+	// --- parseProbabilities tests ---
+
+	@Test
+	public void testProbabilitiesAbsentWhenNoProbsKey() {
+		String json = "{\"content\":\"hi\",\"stop\":true,\"stop_type\":\"eos\"}";
+		LlamaOutput output = parser.parse(json);
+		assertTrue("No completion_probabilities key → empty map", output.probabilities.isEmpty());
+	}
+
+	@Test
+	public void testProbabilitiesParsedPostSampling() {
+		// post_sampling_probs=true → "prob" key
+		String json = "{\"content\":\"hi\",\"stop\":true,\"stop_type\":\"eos\"," +
+				"\"completion_probabilities\":[" +
+				"{\"token\":\"Hello\",\"bytes\":[72],\"id\":15043,\"prob\":0.82," +
+				"\"top_probs\":[{\"token\":\"Hi\",\"bytes\":[72],\"id\":9932,\"prob\":0.1}]}," +
+				"{\"token\":\" world\",\"bytes\":[32,119],\"id\":1917,\"prob\":0.65," +
+				"\"top_probs\":[{\"token\":\" World\",\"bytes\":[32,87],\"id\":2304,\"prob\":0.2}]}" +
+				"]}";
+		LlamaOutput output = parser.parse(json);
+		assertEquals(2, output.probabilities.size());
+		assertEquals(0.82f, output.probabilities.get("Hello"), 0.001f);
+		assertEquals(0.65f, output.probabilities.get(" world"), 0.001f);
+	}
+
+	@Test
+	public void testProbabilitiesParsedPreSampling() {
+		// post_sampling_probs=false → "logprob" key
+		String json = "{\"content\":\"hi\",\"stop\":true,\"stop_type\":\"eos\"," +
+				"\"completion_probabilities\":[" +
+				"{\"token\":\"Hello\",\"bytes\":[72],\"id\":15043,\"logprob\":-0.2," +
+				"\"top_logprobs\":[{\"token\":\"Hi\",\"bytes\":[72],\"id\":9932,\"logprob\":-2.3}]}" +
+				"]}";
+		LlamaOutput output = parser.parse(json);
+		assertEquals(1, output.probabilities.size());
+		assertEquals(-0.2f, output.probabilities.get("Hello"), 0.001f);
+	}
+
+	@Test
+	public void testProbabilitiesTokenWithEscapedChars() {
+		String json = "{\"content\":\"hi\",\"stop\":true,\"stop_type\":\"eos\"," +
+				"\"completion_probabilities\":[" +
+				"{\"token\":\"say \\\"yes\\\"\",\"bytes\":[],\"id\":1,\"prob\":0.5," +
+				"\"top_probs\":[]}" +
+				"]}";
+		LlamaOutput output = parser.parse(json);
+		assertEquals(1, output.probabilities.size());
+		assertEquals(0.5f, output.probabilities.get("say \"yes\""), 0.001f);
+	}
+
+	// --- StopReason tests ---
+
+	@Test
+	public void testStopReasonNoneOnIntermediateToken() {
+		LlamaOutput output = new LlamaOutput("token", Collections.emptyMap(), false, StopReason.NONE);
+		assertEquals(StopReason.NONE, output.stopReason);
+	}
+
+	@Test
+	public void testStopReasonFromJsonEos() {
+		String json = "{\"content\":\"done\",\"stop\":true,\"stop_type\":\"eos\"}";
+		LlamaOutput output = parser.parse(json);
+		assertTrue(output.stop);
+		assertEquals(StopReason.EOS, output.stopReason);
+	}
+
+	@Test
+	public void testStopReasonFromJsonWord() {
+		String json = "{\"content\":\"done\",\"stop\":true,\"stop_type\":\"word\",\"stopping_word\":\"END\"}";
+		LlamaOutput output = parser.parse(json);
+		assertTrue(output.stop);
+		assertEquals(StopReason.STOP_STRING, output.stopReason);
+	}
+
+	@Test
+	public void testStopReasonFromJsonLimit() {
+		String json = "{\"content\":\"truncated\",\"stop\":true,\"stop_type\":\"limit\",\"truncated\":true}";
+		LlamaOutput output = parser.parse(json);
+		assertTrue(output.stop);
+		assertEquals(StopReason.MAX_TOKENS, output.stopReason);
+	}
+
+	@Test
+	public void testStopReasonNoneWhenStopFalse() {
+		String json = "{\"content\":\"partial\",\"stop\":false,\"stop_type\":\"eos\"}";
+		LlamaOutput output = parser.parse(json);
+		assertFalse(output.stop);
+		// stopReason is NONE for non-final tokens regardless of stop_type
+		assertEquals(StopReason.NONE, output.stopReason);
 	}
 }
diff --git a/src/test/java/de/kherud/llama/ModelMetaTest.java b/src/test/java/de/kherud/llama/ModelMetaTest.java
new file mode 100644
index 00000000..0b4b87ee
--- /dev/null
+++ b/src/test/java/de/kherud/llama/ModelMetaTest.java
@@ -0,0 +1,118 @@
+package de.kherud.llama;
+
+import com.fasterxml.jackson.databind.ObjectMapper;
+import org.junit.Test;
+
+import static org.junit.Assert.*;
+
+/**
+ * Unit tests for {@link ModelMeta} typed getters.
+ * Constructs {@code ModelMeta} directly from JSON strings — no native library or model file required.
+ */
+@ClaudeGenerated(
+        purpose = "Verify that ModelMeta typed getters map correctly from the underlying JsonNode, " +
+                  "including the new architecture and name fields from GGUF general.* metadata."
+)
+public class ModelMetaTest {
+
+    private static final ObjectMapper MAPPER = new ObjectMapper();
+
+    private ModelMeta parse(String json) throws Exception {
+        return new ModelMeta(MAPPER.readTree(json));
+    }
+
+    @Test
+    public void testNumericGetters() throws Exception {
+        ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":32016,\"n_ctx_train\":16384," +
+                "\"n_embd\":4096,\"n_params\":6738546688,\"size\":2825274880," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}," +
+                "\"architecture\":\"llama\",\"name\":\"CodeLlama-7B\"}");
+
+        assertEquals(1, meta.getVocabType());
+        assertEquals(32016, meta.getNVocab());
+        assertEquals(16384, meta.getNCtxTrain());
+        assertEquals(4096, meta.getNEmbd());
+        assertEquals(6738546688L, meta.getNParams());
+        assertEquals(2825274880L, meta.getSize());
+    }
+
+    @Test
+    public void testModalityGetters() throws Exception {
+        ModelMeta textOnly = parse("{\"vocab_type\":1,\"n_vocab\":100,\"n_ctx_train\":4096," +
+                "\"n_embd\":512,\"n_params\":1000000,\"size\":500000," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}," +
+                "\"architecture\":\"llama\",\"name\":\"\"}");
+        assertFalse(textOnly.supportsVision());
+        assertFalse(textOnly.supportsAudio());
+
+        ModelMeta multimodal = parse("{\"vocab_type\":1,\"n_vocab\":100,\"n_ctx_train\":4096," +
+                "\"n_embd\":512,\"n_params\":1000000,\"size\":500000," +
+                "\"modalities\":{\"vision\":true,\"audio\":true}," +
+                "\"architecture\":\"gemma3\",\"name\":\"Gemma-3\"}");
+        assertTrue(multimodal.supportsVision());
+        assertTrue(multimodal.supportsAudio());
+    }
+
+    @Test
+    public void testGetArchitecture() throws Exception {
+        ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":32016,\"n_ctx_train\":16384," +
+                "\"n_embd\":4096,\"n_params\":6738546688,\"size\":2825274880," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}," +
+                "\"architecture\":\"llama\",\"name\":\"CodeLlama-7B\"}");
+
+        assertEquals("llama", meta.getArchitecture());
+    }
+
+    @Test
+    public void testGetModelName() throws Exception {
+        ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":32016,\"n_ctx_train\":16384," +
+                "\"n_embd\":4096,\"n_params\":6738546688,\"size\":2825274880," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}," +
+                "\"architecture\":\"mistral\",\"name\":\"Mistral-7B-v0.1\"}");
+
+        assertEquals("Mistral-7B-v0.1", meta.getModelName());
+    }
+
+    @Test
+    public void testGetArchitectureEmptyWhenAbsent() throws Exception {
+        ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":100,\"n_ctx_train\":4096," +
+                "\"n_embd\":512,\"n_params\":1000000,\"size\":500000," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}}");
+
+        assertEquals("", meta.getArchitecture());
+    }
+
+    @Test
+    public void testGetModelNameEmptyWhenAbsent() throws Exception {
+        ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":100,\"n_ctx_train\":4096," +
+                "\"n_embd\":512,\"n_params\":1000000,\"size\":500000," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}}");
+
+        assertEquals("", meta.getModelName());
+    }
+
+    @Test
+    public void testGetArchitectureVariousModels() throws Exception {
+        for (String arch : new String[]{"llama", "gemma3", "mistral", "falcon", "phi3"}) {
+            ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":100,\"n_ctx_train\":4096," +
+                    "\"n_embd\":512,\"n_params\":1000000,\"size\":500000," +
+                    "\"modalities\":{\"vision\":false,\"audio\":false}," +
+                    "\"architecture\":\"" + arch + "\",\"name\":\"\"}");
+            assertEquals(arch, meta.getArchitecture());
+        }
+    }
+
+    @Test
+    public void testToStringContainsNewFields() throws Exception {
+        ModelMeta meta = parse("{\"vocab_type\":1,\"n_vocab\":32016,\"n_ctx_train\":16384," +
+                "\"n_embd\":4096,\"n_params\":6738546688,\"size\":2825274880," +
+                "\"modalities\":{\"vision\":false,\"audio\":false}," +
+                "\"architecture\":\"llama\",\"name\":\"CodeLlama-7B\"}");
+
+        String json = meta.toString();
+        assertTrue(json.contains("\"architecture\""));
+        assertTrue(json.contains("\"name\""));
+        assertTrue(json.contains("\"llama\""));
+        assertTrue(json.contains("\"CodeLlama-7B\""));
+    }
+}
diff --git a/src/test/java/de/kherud/llama/StopReasonTest.java b/src/test/java/de/kherud/llama/StopReasonTest.java
new file mode 100644
index 00000000..0f8af234
--- /dev/null
+++ b/src/test/java/de/kherud/llama/StopReasonTest.java
@@ -0,0 +1,66 @@
+package de.kherud.llama;
+
+import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
+
+import static org.junit.Assert.*;
+
+/**
+ * Round-trip tests for {@link StopReason}.
+ *
+ * <p>The parameterised suite drives one test per enum constant: it obtains the
+ * constant's {@code "stop_type"} string via {@link StopReason#getStopType()} and
+ * verifies that feeding it back into {@link StopReason#fromStopType(String)} returns
+ * the original constant.  The data provider is {@link StopReason#values()} so the
+ * suite automatically covers any future constant added to the enum.
+ *
+ * <p>Edge cases (null, empty string, unknown value) are tested in separate
+ * {@code @Test} methods below the round-trip test.
+ */
+@RunWith(Parameterized.class)
+public class StopReasonTest {
+
+    @Parameterized.Parameters(name = "{0}")
+    public static Collection<StopReason> data() {
+        return Arrays.asList(StopReason.values());
+    }
+
+    private final StopReason reason;
+
+    public StopReasonTest(StopReason reason) {
+        this.reason = reason;
+    }
+
+    @Test
+    public void testRoundTrip() {
+        assertSame(reason, StopReason.fromStopType(reason.getStopType()));
+    }
+
+    // ------------------------------------------------------------------
+    // Edge cases — tested separately from the round-trip
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testFromStopType_nullReturnsNone() {
+        assertSame(StopReason.NONE, StopReason.fromStopType(null));
+    }
+
+    @Test
+    public void testFromStopType_emptyStringReturnsNone() {
+        assertSame(StopReason.NONE, StopReason.fromStopType(""));
+    }
+
+    @Test
+    public void testFromStopType_unknownReturnsNone() {
+        assertSame(StopReason.NONE, StopReason.fromStopType("something_else"));
+    }
+
+    @Test
+    public void testEnumCount() {
+        assertEquals(4, StopReason.values().length);
+    }
+}
diff --git a/src/test/java/de/kherud/llama/args/CacheTypeTest.java b/src/test/java/de/kherud/llama/args/CacheTypeTest.java
index 76bbee9d..1979db34 100644
--- a/src/test/java/de/kherud/llama/args/CacheTypeTest.java
+++ b/src/test/java/de/kherud/llama/args/CacheTypeTest.java
@@ -1,72 +1,62 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify CacheType enum values, count, and lowercase name convention used by ModelParameters.",
-        model = "claude-opus-4-6"
-)
+@RunWith(Parameterized.class)
 public class CacheTypeTest {
 
-    @Test
-    public void testEnumCount() {
-        assertEquals(9, CacheType.values().length);
-    }
-
-    @Test
-    public void testF32() {
-        assertEquals("f32", CacheType.F32.name().toLowerCase());
-    }
-
-    @Test
-    public void testF16() {
-        assertEquals("f16", CacheType.F16.name().toLowerCase());
-    }
-
-    @Test
-    public void testBF16() {
-        assertEquals("bf16", CacheType.BF16.name().toLowerCase());
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {CacheType.F32,    "f32"},
+            {CacheType.F16,    "f16"},
+            {CacheType.BF16,   "bf16"},
+            {CacheType.Q8_0,   "q8_0"},
+            {CacheType.Q4_0,   "q4_0"},
+            {CacheType.Q4_1,   "q4_1"},
+            {CacheType.IQ4_NL, "iq4_nl"},
+            {CacheType.Q5_0,   "q5_0"},
+            {CacheType.Q5_1,   "q5_1"},
+        });
     }
 
-    @Test
-    public void testQ8_0() {
-        assertEquals("q8_0", CacheType.Q8_0.name().toLowerCase());
-    }
+    private final CacheType cacheType;
+    private final String expectedArgValue;
 
-    @Test
-    public void testQ4_0() {
-        assertEquals("q4_0", CacheType.Q4_0.name().toLowerCase());
+    public CacheTypeTest(CacheType cacheType, String expectedArgValue) {
+        this.cacheType = cacheType;
+        this.expectedArgValue = expectedArgValue;
     }
 
     @Test
-    public void testQ4_1() {
-        assertEquals("q4_1", CacheType.Q4_1.name().toLowerCase());
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, cacheType.getArgValue());
     }
 
-    @Test
-    public void testIQ4_NL() {
-        assertEquals("iq4_nl", CacheType.IQ4_NL.name().toLowerCase());
-    }
+    // ------------------------------------------------------------------
+    // Structural invariants — tested separately from the per-value check
+    // ------------------------------------------------------------------
 
     @Test
-    public void testQ5_0() {
-        assertEquals("q5_0", CacheType.Q5_0.name().toLowerCase());
+    public void testEnumCount() {
+        assertEquals(9, CacheType.values().length);
     }
 
     @Test
-    public void testQ5_1() {
-        assertEquals("q5_1", CacheType.Q5_1.name().toLowerCase());
+    public void testImplementsCliArg() {
+        assertTrue(cacheType instanceof CliArg);
     }
 
     @Test
-    public void testAllValuesHaveNonEmptyLowercaseName() {
-        for (CacheType ct : CacheType.values()) {
-            String lower = ct.name().toLowerCase();
-            assertNotNull(lower);
-            assertFalse("CacheType " + ct + " has empty lowercase name", lower.isEmpty());
-        }
+    public void testArgValueNonEmpty() {
+        assertNotNull(cacheType.getArgValue());
+        assertFalse(cacheType.getArgValue().isEmpty());
     }
 }
diff --git a/src/test/java/de/kherud/llama/args/GpuSplitModeTest.java b/src/test/java/de/kherud/llama/args/GpuSplitModeTest.java
index 429e88da..9e40363c 100644
--- a/src/test/java/de/kherud/llama/args/GpuSplitModeTest.java
+++ b/src/test/java/de/kherud/llama/args/GpuSplitModeTest.java
@@ -1,42 +1,56 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify GpuSplitMode enum values, count, and lowercase name convention used by ModelParameters.",
-        model = "claude-opus-4-6"
-)
+@RunWith(Parameterized.class)
 public class GpuSplitModeTest {
 
-    @Test
-    public void testEnumCount() {
-        assertEquals(3, GpuSplitMode.values().length);
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {GpuSplitMode.NONE,  "none"},
+            {GpuSplitMode.LAYER, "layer"},
+            {GpuSplitMode.ROW,   "row"},
+        });
+    }
+
+    private final GpuSplitMode gpuSplitMode;
+    private final String expectedArgValue;
+
+    public GpuSplitModeTest(GpuSplitMode gpuSplitMode, String expectedArgValue) {
+        this.gpuSplitMode = gpuSplitMode;
+        this.expectedArgValue = expectedArgValue;
     }
 
     @Test
-    public void testNone() {
-        assertEquals("none", GpuSplitMode.NONE.name().toLowerCase());
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, gpuSplitMode.getArgValue());
     }
 
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
+
     @Test
-    public void testLayer() {
-        assertEquals("layer", GpuSplitMode.LAYER.name().toLowerCase());
+    public void testEnumCount() {
+        assertEquals(3, GpuSplitMode.values().length);
     }
 
     @Test
-    public void testRow() {
-        assertEquals("row", GpuSplitMode.ROW.name().toLowerCase());
+    public void testImplementsCliArg() {
+        assertTrue(gpuSplitMode instanceof CliArg);
     }
 
     @Test
-    public void testAllValuesHaveNonEmptyLowercaseName() {
-        for (GpuSplitMode mode : GpuSplitMode.values()) {
-            String lower = mode.name().toLowerCase();
-            assertNotNull(lower);
-            assertFalse("GpuSplitMode " + mode + " has empty lowercase name", lower.isEmpty());
-        }
+    public void testArgValueNonEmpty() {
+        assertNotNull(gpuSplitMode.getArgValue());
+        assertFalse(gpuSplitMode.getArgValue().isEmpty());
     }
 }
diff --git a/src/test/java/de/kherud/llama/args/MiroStatTest.java b/src/test/java/de/kherud/llama/args/MiroStatTest.java
index 5610215e..49a91e89 100644
--- a/src/test/java/de/kherud/llama/args/MiroStatTest.java
+++ b/src/test/java/de/kherud/llama/args/MiroStatTest.java
@@ -1,40 +1,56 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify MiroStat enum values and count.",
-        model = "claude-opus-4-6"
-)
+@RunWith(Parameterized.class)
 public class MiroStatTest {
 
-    @Test
-    public void testEnumCount() {
-        assertEquals(3, MiroStat.values().length);
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {MiroStat.DISABLED, "0"},
+            {MiroStat.V1,       "1"},
+            {MiroStat.V2,       "2"},
+        });
+    }
+
+    private final MiroStat miroStat;
+    private final String expectedArgValue;
+
+    public MiroStatTest(MiroStat miroStat, String expectedArgValue) {
+        this.miroStat = miroStat;
+        this.expectedArgValue = expectedArgValue;
     }
 
     @Test
-    public void testDisabledOrdinal() {
-        assertEquals(0, MiroStat.DISABLED.ordinal());
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, miroStat.getArgValue());
     }
 
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
+
     @Test
-    public void testV1Ordinal() {
-        assertEquals(1, MiroStat.V1.ordinal());
+    public void testEnumCount() {
+        assertEquals(3, MiroStat.values().length);
     }
 
     @Test
-    public void testV2Ordinal() {
-        assertEquals(2, MiroStat.V2.ordinal());
+    public void testImplementsCliArg() {
+        assertTrue(miroStat instanceof CliArg);
     }
 
     @Test
-    public void testValueOf() {
-        assertEquals(MiroStat.DISABLED, MiroStat.valueOf("DISABLED"));
-        assertEquals(MiroStat.V1, MiroStat.valueOf("V1"));
-        assertEquals(MiroStat.V2, MiroStat.valueOf("V2"));
+    public void testArgValueNonEmpty() {
+        assertNotNull(miroStat.getArgValue());
+        assertFalse(miroStat.getArgValue().isEmpty());
     }
 }
diff --git a/src/test/java/de/kherud/llama/args/ModelFlagTest.java b/src/test/java/de/kherud/llama/args/ModelFlagTest.java
new file mode 100644
index 00000000..16ce3e44
--- /dev/null
+++ b/src/test/java/de/kherud/llama/args/ModelFlagTest.java
@@ -0,0 +1,81 @@
+package de.kherud.llama.args;
+
+import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
+
+import static org.junit.Assert.*;
+
+@RunWith(Parameterized.class)
+public class ModelFlagTest {
+
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {ModelFlag.NO_CONTEXT_SHIFT,       "--no-context-shift"},
+            {ModelFlag.FLASH_ATTN,             "--flash-attn"},
+            {ModelFlag.NO_PERF,                "--no-perf"},
+            {ModelFlag.ESCAPE,                 "--escape"},
+            {ModelFlag.NO_ESCAPE,              "--no-escape"},
+            {ModelFlag.SPECIAL,                "--special"},
+            {ModelFlag.NO_WARMUP,              "--no-warmup"},
+            {ModelFlag.SPM_INFILL,             "--spm-infill"},
+            {ModelFlag.IGNORE_EOS,             "--ignore-eos"},
+            {ModelFlag.DUMP_KV_CACHE,          "--dump-kv-cache"},
+            {ModelFlag.NO_KV_OFFLOAD,          "--no-kv-offload"},
+            {ModelFlag.CONT_BATCHING,          "--cont-batching"},
+            {ModelFlag.NO_CONT_BATCHING,       "--no-cont-batching"},
+            {ModelFlag.MLOCK,                  "--mlock"},
+            {ModelFlag.NO_MMAP,                "--no-mmap"},
+            {ModelFlag.CHECK_TENSORS,          "--check-tensors"},
+            {ModelFlag.EMBEDDING,              "--embedding"},
+            {ModelFlag.RERANKING,              "--reranking"},
+            {ModelFlag.LORA_INIT_WITHOUT_APPLY,"--lora-init-without-apply"},
+            {ModelFlag.LOG_DISABLE,            "--log-disable"},
+            {ModelFlag.VERBOSE,                "--verbose"},
+            {ModelFlag.LOG_PREFIX,             "--log-prefix"},
+            {ModelFlag.LOG_TIMESTAMPS,         "--log-timestamps"},
+            {ModelFlag.JINJA,                  "--jinja"},
+            {ModelFlag.VOCAB_ONLY,             "--vocab-only"},
+            {ModelFlag.KV_UNIFIED,             "--kv-unified"},
+            {ModelFlag.NO_KV_UNIFIED,          "--no-kv-unified"},
+            {ModelFlag.CLEAR_IDLE,             "--clear-idle"},
+            {ModelFlag.NO_CLEAR_IDLE,          "--no-clear-idle"},
+        });
+    }
+
+    private final ModelFlag flag;
+    private final String expectedCliFlag;
+
+    public ModelFlagTest(ModelFlag flag, String expectedCliFlag) {
+        this.flag = flag;
+        this.expectedCliFlag = expectedCliFlag;
+    }
+
+    @Test
+    public void testGetCliFlag() {
+        assertEquals(expectedCliFlag, flag.getCliFlag());
+    }
+
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testEnumCount() {
+        assertEquals(29, ModelFlag.values().length);
+    }
+
+    @Test
+    public void testCliFlagStartsWithDoubleDash() {
+        assertTrue("Flag " + flag + " must start with --", flag.getCliFlag().startsWith("--"));
+    }
+
+    @Test
+    public void testCliFlagNonEmpty() {
+        assertFalse("Flag " + flag + " has empty CLI string", flag.getCliFlag().isEmpty());
+    }
+}
diff --git a/src/test/java/de/kherud/llama/args/NumaStrategyTest.java b/src/test/java/de/kherud/llama/args/NumaStrategyTest.java
index 4c3477f3..fd6fcc6f 100644
--- a/src/test/java/de/kherud/llama/args/NumaStrategyTest.java
+++ b/src/test/java/de/kherud/llama/args/NumaStrategyTest.java
@@ -1,42 +1,56 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify NumaStrategy enum values, count, and lowercase name convention used by ModelParameters.",
-        model = "claude-opus-4-6"
-)
+@RunWith(Parameterized.class)
 public class NumaStrategyTest {
 
-    @Test
-    public void testEnumCount() {
-        assertEquals(3, NumaStrategy.values().length);
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {NumaStrategy.DISTRIBUTE, "distribute"},
+            {NumaStrategy.ISOLATE,    "isolate"},
+            {NumaStrategy.NUMACTL,    "numactl"},
+        });
+    }
+
+    private final NumaStrategy numaStrategy;
+    private final String expectedArgValue;
+
+    public NumaStrategyTest(NumaStrategy numaStrategy, String expectedArgValue) {
+        this.numaStrategy = numaStrategy;
+        this.expectedArgValue = expectedArgValue;
     }
 
     @Test
-    public void testDistribute() {
-        assertEquals("distribute", NumaStrategy.DISTRIBUTE.name().toLowerCase());
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, numaStrategy.getArgValue());
     }
 
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
+
     @Test
-    public void testIsolate() {
-        assertEquals("isolate", NumaStrategy.ISOLATE.name().toLowerCase());
+    public void testEnumCount() {
+        assertEquals(3, NumaStrategy.values().length);
     }
 
     @Test
-    public void testNumactl() {
-        assertEquals("numactl", NumaStrategy.NUMACTL.name().toLowerCase());
+    public void testImplementsCliArg() {
+        assertTrue(numaStrategy instanceof CliArg);
     }
 
     @Test
-    public void testAllValuesHaveNonEmptyLowercaseName() {
-        for (NumaStrategy ns : NumaStrategy.values()) {
-            String lower = ns.name().toLowerCase();
-            assertNotNull(lower);
-            assertFalse("NumaStrategy " + ns + " has empty lowercase name", lower.isEmpty());
-        }
+    public void testArgValueNonEmpty() {
+        assertNotNull(numaStrategy.getArgValue());
+        assertFalse(numaStrategy.getArgValue().isEmpty());
     }
 }
diff --git a/src/test/java/de/kherud/llama/args/PoolingTypeTest.java b/src/test/java/de/kherud/llama/args/PoolingTypeTest.java
index 605402bd..1daeb71c 100644
--- a/src/test/java/de/kherud/llama/args/PoolingTypeTest.java
+++ b/src/test/java/de/kherud/llama/args/PoolingTypeTest.java
@@ -1,57 +1,59 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify that every PoolingType enum constant returns the exact CLI argument " +
-                  "string expected by llama.cpp (e.g. MEAN -> \"mean\", RANK -> \"rank\") via " +
-                  "getArgValue(), and that the enum has the expected number of constants."
-)
+@RunWith(Parameterized.class)
 public class PoolingTypeTest {
 
-	@Test
-	public void testUnspecifiedArgValue() {
-		assertEquals("unspecified", PoolingType.UNSPECIFIED.getArgValue());
-	}
-
-	@Test
-	public void testNoneArgValue() {
-		assertEquals("none", PoolingType.NONE.getArgValue());
-	}
-
-	@Test
-	public void testMeanArgValue() {
-		assertEquals("mean", PoolingType.MEAN.getArgValue());
-	}
-
-	@Test
-	public void testClsArgValue() {
-		assertEquals("cls", PoolingType.CLS.getArgValue());
-	}
-
-	@Test
-	public void testLastArgValue() {
-		assertEquals("last", PoolingType.LAST.getArgValue());
-	}
-
-	@Test
-	public void testRankArgValue() {
-		assertEquals("rank", PoolingType.RANK.getArgValue());
-	}
-
-	@Test
-	public void testAllValuesHaveArgValue() {
-		for (PoolingType type : PoolingType.values()) {
-			assertNotNull("getArgValue() should not be null for " + type, type.getArgValue());
-			assertFalse("getArgValue() should not be empty for " + type, type.getArgValue().isEmpty());
-		}
-	}
-
-	@Test
-	public void testEnumCount() {
-		assertEquals(6, PoolingType.values().length);
-	}
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {PoolingType.UNSPECIFIED, "unspecified"},
+            {PoolingType.NONE,        "none"},
+            {PoolingType.MEAN,        "mean"},
+            {PoolingType.CLS,         "cls"},
+            {PoolingType.LAST,        "last"},
+            {PoolingType.RANK,        "rank"},
+        });
+    }
+
+    private final PoolingType poolingType;
+    private final String expectedArgValue;
+
+    public PoolingTypeTest(PoolingType poolingType, String expectedArgValue) {
+        this.poolingType = poolingType;
+        this.expectedArgValue = expectedArgValue;
+    }
+
+    @Test
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, poolingType.getArgValue());
+    }
+
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testEnumCount() {
+        assertEquals(6, PoolingType.values().length);
+    }
+
+    @Test
+    public void testImplementsCliArg() {
+        assertTrue(poolingType instanceof CliArg);
+    }
+
+    @Test
+    public void testArgValueNonEmpty() {
+        assertNotNull(poolingType.getArgValue());
+        assertFalse(poolingType.getArgValue().isEmpty());
+    }
 }
diff --git a/src/test/java/de/kherud/llama/args/RopeScalingTypeTest.java b/src/test/java/de/kherud/llama/args/RopeScalingTypeTest.java
index fce82846..e1635a8b 100644
--- a/src/test/java/de/kherud/llama/args/RopeScalingTypeTest.java
+++ b/src/test/java/de/kherud/llama/args/RopeScalingTypeTest.java
@@ -1,57 +1,59 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify that every RopeScalingType enum constant returns the exact CLI argument " +
-                  "string expected by llama.cpp (e.g. YARN2 -> \"yarn\", LONGROPE -> \"longrope\") " +
-                  "via getArgValue(), and that the enum has the expected number of constants."
-)
+@RunWith(Parameterized.class)
 public class RopeScalingTypeTest {
 
-	@Test
-	public void testUnspecifiedArgValue() {
-		assertEquals("unspecified", RopeScalingType.UNSPECIFIED.getArgValue());
-	}
-
-	@Test
-	public void testNoneArgValue() {
-		assertEquals("none", RopeScalingType.NONE.getArgValue());
-	}
-
-	@Test
-	public void testLinearArgValue() {
-		assertEquals("linear", RopeScalingType.LINEAR.getArgValue());
-	}
-
-	@Test
-	public void testYarn2ArgValue() {
-		assertEquals("yarn", RopeScalingType.YARN2.getArgValue());
-	}
-
-	@Test
-	public void testLongRopeArgValue() {
-		assertEquals("longrope", RopeScalingType.LONGROPE.getArgValue());
-	}
-
-	@Test
-	public void testMaxValueArgValue() {
-		assertEquals("maxvalue", RopeScalingType.MAX_VALUE.getArgValue());
-	}
-
-	@Test
-	public void testAllValuesHaveArgValue() {
-		for (RopeScalingType type : RopeScalingType.values()) {
-			assertNotNull("getArgValue() should not be null for " + type, type.getArgValue());
-			assertFalse("getArgValue() should not be empty for " + type, type.getArgValue().isEmpty());
-		}
-	}
-
-	@Test
-	public void testEnumCount() {
-		assertEquals(6, RopeScalingType.values().length);
-	}
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {RopeScalingType.UNSPECIFIED, "unspecified"},
+            {RopeScalingType.NONE,        "none"},
+            {RopeScalingType.LINEAR,      "linear"},
+            {RopeScalingType.YARN2,       "yarn"},
+            {RopeScalingType.LONGROPE,    "longrope"},
+            {RopeScalingType.MAX_VALUE,   "maxvalue"},
+        });
+    }
+
+    private final RopeScalingType ropeScalingType;
+    private final String expectedArgValue;
+
+    public RopeScalingTypeTest(RopeScalingType ropeScalingType, String expectedArgValue) {
+        this.ropeScalingType = ropeScalingType;
+        this.expectedArgValue = expectedArgValue;
+    }
+
+    @Test
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, ropeScalingType.getArgValue());
+    }
+
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testEnumCount() {
+        assertEquals(6, RopeScalingType.values().length);
+    }
+
+    @Test
+    public void testImplementsCliArg() {
+        assertTrue(ropeScalingType instanceof CliArg);
+    }
+
+    @Test
+    public void testArgValueNonEmpty() {
+        assertNotNull(ropeScalingType.getArgValue());
+        assertFalse(ropeScalingType.getArgValue().isEmpty());
+    }
 }
diff --git a/src/test/java/de/kherud/llama/args/SamplerTest.java b/src/test/java/de/kherud/llama/args/SamplerTest.java
index 846c6667..b0518af8 100644
--- a/src/test/java/de/kherud/llama/args/SamplerTest.java
+++ b/src/test/java/de/kherud/llama/args/SamplerTest.java
@@ -1,73 +1,62 @@
 package de.kherud.llama.args;
 
-import de.kherud.llama.ClaudeGenerated;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
+
+import java.util.Arrays;
+import java.util.Collection;
 
 import static org.junit.Assert.*;
 
-@ClaudeGenerated(
-        purpose = "Verify Sampler enum values, count, and lowercase name convention used by " +
-                  "ModelParameters.setSamplers() semicolon-separated serialization.",
-        model = "claude-opus-4-6"
-)
+@RunWith(Parameterized.class)
 public class SamplerTest {
 
-    @Test
-    public void testEnumCount() {
-        assertEquals(9, Sampler.values().length);
-    }
-
-    @Test
-    public void testDry() {
-        assertEquals("dry", Sampler.DRY.name().toLowerCase());
-    }
-
-    @Test
-    public void testTopK() {
-        assertEquals("top_k", Sampler.TOP_K.name().toLowerCase());
-    }
-
-    @Test
-    public void testTopP() {
-        assertEquals("top_p", Sampler.TOP_P.name().toLowerCase());
+    @Parameterized.Parameters(name = "{0} -> {1}")
+    public static Collection<Object[]> data() {
+        return Arrays.asList(new Object[][]{
+            {Sampler.DRY,         "dry"},
+            {Sampler.TOP_K,       "top_k"},
+            {Sampler.TOP_P,       "top_p"},
+            {Sampler.TYP_P,       "typ_p"},
+            {Sampler.MIN_P,       "min_p"},
+            {Sampler.TEMPERATURE, "temperature"},
+            {Sampler.XTC,         "xtc"},
+            {Sampler.INFILL,      "infill"},
+            {Sampler.PENALTIES,   "penalties"},
+        });
     }
 
-    @Test
-    public void testTypP() {
-        assertEquals("typ_p", Sampler.TYP_P.name().toLowerCase());
-    }
+    private final Sampler sampler;
+    private final String expectedArgValue;
 
-    @Test
-    public void testMinP() {
-        assertEquals("min_p", Sampler.MIN_P.name().toLowerCase());
+    public SamplerTest(Sampler sampler, String expectedArgValue) {
+        this.sampler = sampler;
+        this.expectedArgValue = expectedArgValue;
     }
 
     @Test
-    public void testTemperature() {
-        assertEquals("temperature", Sampler.TEMPERATURE.name().toLowerCase());
+    public void testGetArgValue() {
+        assertEquals(expectedArgValue, sampler.getArgValue());
     }
 
-    @Test
-    public void testXtc() {
-        assertEquals("xtc", Sampler.XTC.name().toLowerCase());
-    }
+    // ------------------------------------------------------------------
+    // Structural invariants
+    // ------------------------------------------------------------------
 
     @Test
-    public void testInfill() {
-        assertEquals("infill", Sampler.INFILL.name().toLowerCase());
+    public void testEnumCount() {
+        assertEquals(9, Sampler.values().length);
     }
 
     @Test
-    public void testPenalties() {
-        assertEquals("penalties", Sampler.PENALTIES.name().toLowerCase());
+    public void testImplementsCliArg() {
+        assertTrue(sampler instanceof CliArg);
     }
 
     @Test
-    public void testAllValuesHaveNonEmptyLowercaseName() {
-        for (Sampler s : Sampler.values()) {
-            String lower = s.name().toLowerCase();
-            assertNotNull(lower);
-            assertFalse("Sampler " + s + " has empty lowercase name", lower.isEmpty());
-        }
+    public void testArgValueNonEmpty() {
+        assertNotNull(sampler.getArgValue());
+        assertFalse(sampler.getArgValue().isEmpty());
     }
 }
diff --git a/src/test/java/de/kherud/llama/json/ChatResponseParserTest.java b/src/test/java/de/kherud/llama/json/ChatResponseParserTest.java
new file mode 100644
index 00000000..69572862
--- /dev/null
+++ b/src/test/java/de/kherud/llama/json/ChatResponseParserTest.java
@@ -0,0 +1,163 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import org.junit.Test;
+
+import static org.junit.Assert.*;
+
+/**
+ * Unit tests for {@link ChatResponseParser}.
+ * No JVM native library or model file needed — JSON string literals only.
+ */
+public class ChatResponseParserTest {
+
+    private static final ObjectMapper MAPPER = new ObjectMapper();
+    private final ChatResponseParser parser = new ChatResponseParser();
+
+    // ------------------------------------------------------------------
+    // extractChoiceContent(String)
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testExtractChoiceContent_typical() {
+        String json = "{\"choices\":[{\"index\":0,\"message\":{\"role\":\"assistant\",\"content\":\"OK\"}," +
+                "\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":5,\"completion_tokens\":1}}";
+        assertEquals("OK", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_emptyContent() {
+        String json = "{\"choices\":[{\"message\":{\"role\":\"assistant\",\"content\":\"\"}}]}";
+        assertEquals("", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_escapedContent() {
+        String json = "{\"choices\":[{\"message\":{\"role\":\"assistant\"," +
+                "\"content\":\"line1\\nline2\\t\\\"quoted\\\"\"}}]}";
+        assertEquals("line1\nline2\t\"quoted\"", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_unicodeInContent() {
+        String json = "{\"choices\":[{\"message\":{\"role\":\"assistant\",\"content\":\"caf\\u00e9\"}}]}";
+        assertEquals("café", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_missingChoices() {
+        String json = "{\"id\":\"x\",\"object\":\"chat.completion\"}";
+        assertEquals("", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_emptyChoicesArray() {
+        String json = "{\"choices\":[]}";
+        assertEquals("", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_missingContent() {
+        String json = "{\"choices\":[{\"message\":{\"role\":\"assistant\"}}]}";
+        assertEquals("", parser.extractChoiceContent(json));
+    }
+
+    @Test
+    public void testExtractChoiceContent_malformedJson() {
+        assertEquals("", parser.extractChoiceContent("{not json"));
+    }
+
+    @Test
+    public void testExtractChoiceContent_multilineResponse() {
+        String content = "First line.\\nSecond line.\\nThird line.";
+        String json = "{\"choices\":[{\"message\":{\"role\":\"assistant\",\"content\":\"" + content + "\"}}]}";
+        assertEquals("First line.\nSecond line.\nThird line.", parser.extractChoiceContent(json));
+    }
+
+    // ------------------------------------------------------------------
+    // extractChoiceContent(JsonNode)
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testExtractChoiceContent_node() throws Exception {
+        JsonNode node = MAPPER.readTree(
+                "{\"choices\":[{\"message\":{\"role\":\"assistant\",\"content\":\"Hello\"}}]}");
+        assertEquals("Hello", parser.extractChoiceContent(node));
+    }
+
+    @Test
+    public void testExtractChoiceContent_nodeMultipleChoices_takesFirst() throws Exception {
+        JsonNode node = MAPPER.readTree(
+                "{\"choices\":[" +
+                        "{\"message\":{\"content\":\"First\"}}," +
+                        "{\"message\":{\"content\":\"Second\"}}" +
+                        "]}");
+        assertEquals("First", parser.extractChoiceContent(node));
+    }
+
+    // ------------------------------------------------------------------
+    // extractUsageField
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testExtractUsageField_promptTokens() throws Exception {
+        JsonNode node = MAPPER.readTree(
+                "{\"usage\":{\"prompt_tokens\":12,\"completion_tokens\":5,\"total_tokens\":17}}");
+        assertEquals(12, parser.extractUsageField(node, "prompt_tokens"));
+    }
+
+    @Test
+    public void testExtractUsageField_completionTokens() throws Exception {
+        JsonNode node = MAPPER.readTree(
+                "{\"usage\":{\"prompt_tokens\":12,\"completion_tokens\":5,\"total_tokens\":17}}");
+        assertEquals(5, parser.extractUsageField(node, "completion_tokens"));
+    }
+
+    @Test
+    public void testExtractUsageField_totalTokens() throws Exception {
+        JsonNode node = MAPPER.readTree(
+                "{\"usage\":{\"prompt_tokens\":12,\"completion_tokens\":5,\"total_tokens\":17}}");
+        assertEquals(17, parser.extractUsageField(node, "total_tokens"));
+    }
+
+    @Test
+    public void testExtractUsageField_missingUsage_returnsZero() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"id\":\"x\"}");
+        assertEquals(0, parser.extractUsageField(node, "prompt_tokens"));
+    }
+
+    @Test
+    public void testExtractUsageField_missingField_returnsZero() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"usage\":{}}");
+        assertEquals(0, parser.extractUsageField(node, "prompt_tokens"));
+    }
+
+    // ------------------------------------------------------------------
+    // countChoices
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testCountChoices_one() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"choices\":[{\"message\":{\"content\":\"hi\"}}]}");
+        assertEquals(1, parser.countChoices(node));
+    }
+
+    @Test
+    public void testCountChoices_multiple() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"choices\":[{},{},{}]}");
+        assertEquals(3, parser.countChoices(node));
+    }
+
+    @Test
+    public void testCountChoices_empty() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"choices\":[]}");
+        assertEquals(0, parser.countChoices(node));
+    }
+
+    @Test
+    public void testCountChoices_absent() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"id\":\"x\"}");
+        assertEquals(0, parser.countChoices(node));
+    }
+}
diff --git a/src/test/java/de/kherud/llama/json/CompletionResponseParserTest.java b/src/test/java/de/kherud/llama/json/CompletionResponseParserTest.java
new file mode 100644
index 00000000..812c381b
--- /dev/null
+++ b/src/test/java/de/kherud/llama/json/CompletionResponseParserTest.java
@@ -0,0 +1,205 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import de.kherud.llama.LlamaOutput;
+import de.kherud.llama.StopReason;
+import org.junit.Test;
+
+import java.util.Map;
+
+import static org.junit.Assert.*;
+
+/**
+ * Unit tests for {@link CompletionResponseParser}.
+ *
+ * <p>All tests use JSON string literals — no JVM native library or model file is needed.
+ * This mirrors the pattern established by {@code test_json_helpers.cpp} on the C++ side.
+ */
+public class CompletionResponseParserTest {
+
+    private static final ObjectMapper MAPPER = new ObjectMapper();
+    private final CompletionResponseParser parser = new CompletionResponseParser();
+
+    // ------------------------------------------------------------------
+    // parse(String)
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testParseString_text() throws Exception {
+        String json = "{\"content\":\"Hello world\",\"stop\":false}";
+        LlamaOutput out = parser.parse(json);
+        assertEquals("Hello world", out.text);
+    }
+
+    @Test
+    public void testParseString_stopFalse() {
+        String json = "{\"content\":\"partial\",\"stop\":false}";
+        LlamaOutput out = parser.parse(json);
+        assertFalse(out.stop);
+        assertEquals(StopReason.NONE, out.stopReason);
+    }
+
+    @Test
+    public void testParseString_stopTrueEos() {
+        String json = "{\"content\":\"done\",\"stop\":true,\"stop_type\":\"eos\"}";
+        LlamaOutput out = parser.parse(json);
+        assertTrue(out.stop);
+        assertEquals(StopReason.EOS, out.stopReason);
+    }
+
+    @Test
+    public void testParseString_stopTrueWord() {
+        String json = "{\"content\":\"end\",\"stop\":true,\"stop_type\":\"word\",\"stopping_word\":\"END\"}";
+        LlamaOutput out = parser.parse(json);
+        assertTrue(out.stop);
+        assertEquals(StopReason.STOP_STRING, out.stopReason);
+    }
+
+    @Test
+    public void testParseString_stopTrueLimit() {
+        String json = "{\"content\":\"truncated\",\"stop\":true,\"stop_type\":\"limit\",\"truncated\":true}";
+        LlamaOutput out = parser.parse(json);
+        assertTrue(out.stop);
+        assertEquals(StopReason.MAX_TOKENS, out.stopReason);
+    }
+
+    @Test
+    public void testParseString_malformedReturnsEmptyNonStop() {
+        LlamaOutput out = parser.parse("{not valid json");
+        assertEquals("", out.text);
+        assertFalse(out.stop);
+        assertEquals(StopReason.NONE, out.stopReason);
+        assertTrue(out.probabilities.isEmpty());
+    }
+
+    @Test
+    public void testParseString_escapedContent() {
+        String json = "{\"content\":\"line1\\nline2\\t\\\"quoted\\\"\",\"stop\":false}";
+        LlamaOutput out = parser.parse(json);
+        assertEquals("line1\nline2\t\"quoted\"", out.text);
+    }
+
+    @Test
+    public void testParseString_unicodeEscape() {
+        String json = "{\"content\":\"caf\\u00e9\",\"stop\":false}";
+        LlamaOutput out = parser.parse(json);
+        assertEquals("café", out.text);
+    }
+
+    @Test
+    public void testParseString_emptyContent() {
+        String json = "{\"content\":\"\",\"stop\":true,\"stop_type\":\"eos\"}";
+        LlamaOutput out = parser.parse(json);
+        assertEquals("", out.text);
+        assertTrue(out.stop);
+    }
+
+    // ------------------------------------------------------------------
+    // parse(JsonNode)
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testParseNode_delegatesCorrectly() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"content\":\"hi\",\"stop\":true,\"stop_type\":\"eos\"}");
+        LlamaOutput out = parser.parse(node);
+        assertEquals("hi", out.text);
+        assertTrue(out.stop);
+        assertEquals(StopReason.EOS, out.stopReason);
+    }
+
+    // ------------------------------------------------------------------
+    // extractContent
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testExtractContent_present() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"content\":\"hello\",\"stop\":false}");
+        assertEquals("hello", parser.extractContent(node));
+    }
+
+    @Test
+    public void testExtractContent_absent() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"stop\":false}");
+        assertEquals("", parser.extractContent(node));
+    }
+
+    @Test
+    public void testExtractContent_empty() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"content\":\"\",\"stop\":true}");
+        assertEquals("", parser.extractContent(node));
+    }
+
+    // ------------------------------------------------------------------
+    // parseProbabilities
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testParseProbabilities_absentKey() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"content\":\"hi\",\"stop\":true}");
+        assertTrue(parser.parseProbabilities(node).isEmpty());
+    }
+
+    @Test
+    public void testParseProbabilities_emptyArray() throws Exception {
+        JsonNode node = MAPPER.readTree("{\"content\":\"hi\",\"stop\":true,\"completion_probabilities\":[]}");
+        assertTrue(parser.parseProbabilities(node).isEmpty());
+    }
+
+    @Test
+    public void testParseProbabilities_postSampling() throws Exception {
+        String json = "{\"content\":\"hi\",\"stop\":true," +
+                "\"completion_probabilities\":[" +
+                "{\"token\":\"Hello\",\"bytes\":[72],\"id\":15043,\"prob\":0.82," +
+                "\"top_probs\":[{\"token\":\"Hi\",\"bytes\":[72],\"id\":9932,\"prob\":0.1}]}," +
+                "{\"token\":\" world\",\"bytes\":[32,119],\"id\":1917,\"prob\":0.65," +
+                "\"top_probs\":[]}" +
+                "]}";
+        JsonNode node = MAPPER.readTree(json);
+        Map<String, Float> probs = parser.parseProbabilities(node);
+        assertEquals(2, probs.size());
+        assertEquals(0.82f, probs.get("Hello"), 0.001f);
+        assertEquals(0.65f, probs.get(" world"), 0.001f);
+    }
+
+    @Test
+    public void testParseProbabilities_preSampling() throws Exception {
+        String json = "{\"content\":\"hi\",\"stop\":true," +
+                "\"completion_probabilities\":[" +
+                "{\"token\":\"Hello\",\"bytes\":[72],\"id\":15043,\"logprob\":-0.2," +
+                "\"top_logprobs\":[{\"token\":\"Hi\",\"bytes\":[72],\"id\":9932,\"logprob\":-2.3}]}" +
+                "]}";
+        JsonNode node = MAPPER.readTree(json);
+        Map<String, Float> probs = parser.parseProbabilities(node);
+        assertEquals(1, probs.size());
+        assertEquals(-0.2f, probs.get("Hello"), 0.001f);
+    }
+
+    @Test
+    public void testParseProbabilities_escapedToken() throws Exception {
+        String json = "{\"content\":\"hi\",\"stop\":true," +
+                "\"completion_probabilities\":[" +
+                "{\"token\":\"say \\\"yes\\\"\",\"bytes\":[],\"id\":1,\"prob\":0.5," +
+                "\"top_probs\":[]}" +
+                "]}";
+        JsonNode node = MAPPER.readTree(json);
+        Map<String, Float> probs = parser.parseProbabilities(node);
+        assertEquals(1, probs.size());
+        assertEquals(0.5f, probs.get("say \"yes\""), 0.001f);
+    }
+
+    @Test
+    public void testParseProbabilities_topProbs_notIncluded() throws Exception {
+        // top_probs entries must NOT appear in the outer map — only the outer token/prob
+        String json = "{\"content\":\"hi\",\"stop\":true," +
+                "\"completion_probabilities\":[" +
+                "{\"token\":\"A\",\"bytes\":[],\"id\":1,\"prob\":0.9," +
+                "\"top_probs\":[{\"token\":\"B\",\"bytes\":[],\"id\":2,\"prob\":0.05}]}" +
+                "]}";
+        JsonNode node = MAPPER.readTree(json);
+        Map<String, Float> probs = parser.parseProbabilities(node);
+        assertEquals(1, probs.size());
+        assertTrue("only outer token 'A' should be present", probs.containsKey("A"));
+        assertFalse("inner top_probs token 'B' must not appear", probs.containsKey("B"));
+    }
+}
diff --git a/src/test/java/de/kherud/llama/json/ParameterJsonSerializerTest.java b/src/test/java/de/kherud/llama/json/ParameterJsonSerializerTest.java
new file mode 100644
index 00000000..e97e7225
--- /dev/null
+++ b/src/test/java/de/kherud/llama/json/ParameterJsonSerializerTest.java
@@ -0,0 +1,336 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ArrayNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import de.kherud.llama.Pair;
+import de.kherud.llama.args.Sampler;
+import org.junit.Test;
+
+import java.util.Arrays;
+import java.util.Collections;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+
+import static org.junit.Assert.*;
+
+/**
+ * Unit tests for {@link ParameterJsonSerializer}.
+ * No JVM native library or model file needed — plain Java values only.
+ */
+public class ParameterJsonSerializerTest {
+
+    private final ParameterJsonSerializer serializer = new ParameterJsonSerializer();
+
+    // ------------------------------------------------------------------
+    // toJsonString
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testToJsonString_simple() {
+        assertEquals("\"hello\"", serializer.toJsonString("hello"));
+    }
+
+    @Test
+    public void testToJsonString_null() {
+        assertEquals("null", serializer.toJsonString(null));
+    }
+
+    @Test
+    public void testToJsonString_emptyString() {
+        assertEquals("\"\"", serializer.toJsonString(""));
+    }
+
+    @Test
+    public void testToJsonString_newline() {
+        assertEquals("\"line1\\nline2\"", serializer.toJsonString("line1\nline2"));
+    }
+
+    @Test
+    public void testToJsonString_tab() {
+        assertEquals("\"a\\tb\"", serializer.toJsonString("a\tb"));
+    }
+
+    @Test
+    public void testToJsonString_quote() {
+        assertEquals("\"say \\\"hi\\\"\"", serializer.toJsonString("say \"hi\""));
+    }
+
+    @Test
+    public void testToJsonString_backslash() {
+        assertEquals("\"path\\\\file\"", serializer.toJsonString("path\\file"));
+    }
+
+    @Test
+    public void testToJsonString_unicode() {
+        assertEquals("\"café\"", serializer.toJsonString("café"));
+    }
+
+    // ------------------------------------------------------------------
+    // buildMessages
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildMessages_withSystemMessage() {
+        List<Pair<String, String>> msgs = Collections.singletonList(new Pair<>("user", "Hello"));
+        ArrayNode arr = serializer.buildMessages("You are helpful.", msgs);
+        assertEquals(2, arr.size());
+        assertEquals("system", arr.get(0).path("role").asText());
+        assertEquals("You are helpful.", arr.get(0).path("content").asText());
+        assertEquals("user", arr.get(1).path("role").asText());
+        assertEquals("Hello", arr.get(1).path("content").asText());
+    }
+
+    @Test
+    public void testBuildMessages_withoutSystemMessage() {
+        List<Pair<String, String>> msgs = Arrays.asList(
+                new Pair<>("user", "Hi"),
+                new Pair<>("assistant", "Hello there")
+        );
+        ArrayNode arr = serializer.buildMessages(null, msgs);
+        assertEquals(2, arr.size());
+        assertEquals("user", arr.get(0).path("role").asText());
+        assertEquals("assistant", arr.get(1).path("role").asText());
+    }
+
+    @Test
+    public void testBuildMessages_emptySystemMessage_skipped() {
+        List<Pair<String, String>> msgs = Collections.singletonList(new Pair<>("user", "Hi"));
+        ArrayNode arr = serializer.buildMessages("", msgs);
+        assertEquals(1, arr.size());
+        assertEquals("user", arr.get(0).path("role").asText());
+    }
+
+    @Test
+    public void testBuildMessages_specialCharsInContent() {
+        List<Pair<String, String>> msgs = Collections.singletonList(
+                new Pair<>("user", "line1\nline2\t\"quoted\""));
+        ArrayNode arr = serializer.buildMessages(null, msgs);
+        assertEquals("line1\nline2\t\"quoted\"", arr.get(0).path("content").asText());
+    }
+
+    @Test(expected = IllegalArgumentException.class)
+    public void testBuildMessages_invalidRole_throws() {
+        List<Pair<String, String>> msgs = Collections.singletonList(new Pair<>("system", "oops"));
+        serializer.buildMessages(null, msgs);
+    }
+
+    @Test
+    public void testBuildMessages_roundtripsAsJson() throws Exception {
+        List<Pair<String, String>> msgs = Collections.singletonList(new Pair<>("user", "Hello"));
+        ArrayNode arr = serializer.buildMessages("Sys", msgs);
+        String json = arr.toString();
+        JsonNode parsed = serializer.OBJECT_MAPPER.readTree(json);
+        assertEquals("system", parsed.get(0).path("role").asText());
+        assertEquals("Sys", parsed.get(0).path("content").asText());
+        assertEquals("user", parsed.get(1).path("role").asText());
+        assertEquals("Hello", parsed.get(1).path("content").asText());
+    }
+
+    // ------------------------------------------------------------------
+    // buildStopStrings
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildStopStrings_single() {
+        ArrayNode arr = serializer.buildStopStrings("<|endoftext|>");
+        assertEquals(1, arr.size());
+        assertEquals("<|endoftext|>", arr.get(0).asText());
+    }
+
+    @Test
+    public void testBuildStopStrings_multiple() {
+        ArrayNode arr = serializer.buildStopStrings("stop1", "stop2", "stop3");
+        assertEquals(3, arr.size());
+        assertEquals("stop1", arr.get(0).asText());
+        assertEquals("stop3", arr.get(2).asText());
+    }
+
+    @Test
+    public void testBuildStopStrings_withSpecialChars() {
+        ArrayNode arr = serializer.buildStopStrings("line\nnewline", "tab\there");
+        assertEquals("line\nnewline", arr.get(0).asText());
+        assertEquals("tab\there", arr.get(1).asText());
+    }
+
+    @Test
+    public void testBuildStopStrings_roundtripsAsJson() throws Exception {
+        ArrayNode arr = serializer.buildStopStrings("a", "b");
+        JsonNode parsed = serializer.OBJECT_MAPPER.readTree(arr.toString());
+        assertTrue(parsed.isArray());
+        assertEquals("a", parsed.get(0).asText());
+    }
+
+    // ------------------------------------------------------------------
+    // buildSamplers
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildSamplers_allTypes() {
+        ArrayNode arr = serializer.buildSamplers(
+                Sampler.TOP_K, Sampler.TOP_P, Sampler.MIN_P, Sampler.TEMPERATURE);
+        assertEquals(4, arr.size());
+        assertEquals("top_k", arr.get(0).asText());
+        assertEquals("top_p", arr.get(1).asText());
+        assertEquals("min_p", arr.get(2).asText());
+        assertEquals("temperature", arr.get(3).asText());
+    }
+
+    @Test
+    public void testBuildSamplers_single() {
+        ArrayNode arr = serializer.buildSamplers(Sampler.TEMPERATURE);
+        assertEquals(1, arr.size());
+        assertEquals("temperature", arr.get(0).asText());
+    }
+
+    // ------------------------------------------------------------------
+    // buildIntArray
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildIntArray_values() {
+        ArrayNode arr = serializer.buildIntArray(new int[]{1, 2, 3});
+        assertEquals(3, arr.size());
+        assertEquals(1, arr.get(0).asInt());
+        assertEquals(3, arr.get(2).asInt());
+    }
+
+    @Test
+    public void testBuildIntArray_empty() {
+        ArrayNode arr = serializer.buildIntArray(new int[]{});
+        assertEquals(0, arr.size());
+    }
+
+    @Test
+    public void testBuildIntArray_roundtripsAsJson() throws Exception {
+        ArrayNode arr = serializer.buildIntArray(new int[]{10, 20});
+        JsonNode parsed = serializer.OBJECT_MAPPER.readTree(arr.toString());
+        assertTrue(parsed.isArray());
+        assertEquals(10, parsed.get(0).asInt());
+    }
+
+    // ------------------------------------------------------------------
+    // buildTokenIdBiasArray
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildTokenIdBiasArray_structure() {
+        Map<Integer, Float> biases = new LinkedHashMap<>();
+        biases.put(15043, 1.0f);
+        biases.put(50256, -0.5f);
+        ArrayNode arr = serializer.buildTokenIdBiasArray(biases);
+        assertEquals(2, arr.size());
+        assertEquals(15043, arr.get(0).get(0).asInt());
+        assertEquals(1.0, arr.get(0).get(1).asDouble(), 0.001);
+        assertEquals(50256, arr.get(1).get(0).asInt());
+        assertEquals(-0.5, arr.get(1).get(1).asDouble(), 0.001);
+    }
+
+    @Test
+    public void testBuildTokenIdBiasArray_empty() {
+        ArrayNode arr = serializer.buildTokenIdBiasArray(Collections.emptyMap());
+        assertEquals(0, arr.size());
+    }
+
+    // ------------------------------------------------------------------
+    // buildTokenStringBiasArray
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildTokenStringBiasArray_structure() {
+        Map<String, Float> biases = new LinkedHashMap<>();
+        biases.put("Hello", 1.0f);
+        biases.put(" world", -0.5f);
+        ArrayNode arr = serializer.buildTokenStringBiasArray(biases);
+        assertEquals(2, arr.size());
+        assertEquals("Hello", arr.get(0).get(0).asText());
+        assertEquals(1.0, arr.get(0).get(1).asDouble(), 0.001);
+        assertEquals(" world", arr.get(1).get(0).asText());
+    }
+
+    @Test
+    public void testBuildTokenStringBiasArray_specialCharsInKey() {
+        Map<String, Float> biases = new LinkedHashMap<>();
+        biases.put("line\nnewline", 2.0f);
+        ArrayNode arr = serializer.buildTokenStringBiasArray(biases);
+        assertEquals("line\nnewline", arr.get(0).get(0).asText());
+    }
+
+    // ------------------------------------------------------------------
+    // buildDisableTokenIdArray
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildDisableTokenIdArray_structure() {
+        ArrayNode arr = serializer.buildDisableTokenIdArray(Arrays.asList(100, 200, 300));
+        assertEquals(3, arr.size());
+        for (int i = 0; i < arr.size(); i++) {
+            assertFalse(arr.get(i).get(1).asBoolean());
+        }
+        assertEquals(100, arr.get(0).get(0).asInt());
+    }
+
+    @Test
+    public void testBuildDisableTokenIdArray_empty() {
+        ArrayNode arr = serializer.buildDisableTokenIdArray(Collections.emptyList());
+        assertEquals(0, arr.size());
+    }
+
+    // ------------------------------------------------------------------
+    // buildDisableTokenStringArray
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildDisableTokenStringArray_structure() {
+        ArrayNode arr = serializer.buildDisableTokenStringArray(Arrays.asList("foo", "bar"));
+        assertEquals(2, arr.size());
+        assertEquals("foo", arr.get(0).get(0).asText());
+        assertFalse(arr.get(0).get(1).asBoolean());
+        assertEquals("bar", arr.get(1).get(0).asText());
+    }
+
+    // ------------------------------------------------------------------
+    // buildRawValueObject
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testBuildRawValueObject_booleanValue() {
+        Map<String, String> map = Collections.singletonMap("enable_thinking", "true");
+        ObjectNode node = serializer.buildRawValueObject(map);
+        assertTrue(node.path("enable_thinking").isBoolean());
+        assertTrue(node.path("enable_thinking").asBoolean());
+    }
+
+    @Test
+    public void testBuildRawValueObject_numberValue() {
+        Map<String, String> map = Collections.singletonMap("temperature", "0.7");
+        ObjectNode node = serializer.buildRawValueObject(map);
+        assertEquals(0.7, node.path("temperature").asDouble(), 0.001);
+    }
+
+    @Test
+    public void testBuildRawValueObject_stringValue() {
+        Map<String, String> map = Collections.singletonMap("mode", "\"fast\"");
+        ObjectNode node = serializer.buildRawValueObject(map);
+        assertEquals("fast", node.path("mode").asText());
+    }
+
+    @Test
+    public void testBuildRawValueObject_invalidJsonFallsBackToString() {
+        Map<String, String> map = Collections.singletonMap("key", "not-valid-json{{{");
+        ObjectNode node = serializer.buildRawValueObject(map);
+        assertEquals("not-valid-json{{{", node.path("key").asText());
+    }
+
+    @Test
+    public void testBuildRawValueObject_roundtripsAsJson() throws Exception {
+        Map<String, String> map = new LinkedHashMap<>();
+        map.put("flag", "true");
+        map.put("count", "3");
+        ObjectNode node = serializer.buildRawValueObject(map);
+        JsonNode parsed = serializer.OBJECT_MAPPER.readTree(node.toString());
+        assertTrue(parsed.path("flag").asBoolean());
+        assertEquals(3, parsed.path("count").asInt());
+    }
+}
diff --git a/src/test/java/de/kherud/llama/json/RerankResponseParserTest.java b/src/test/java/de/kherud/llama/json/RerankResponseParserTest.java
new file mode 100644
index 00000000..84cc285e
--- /dev/null
+++ b/src/test/java/de/kherud/llama/json/RerankResponseParserTest.java
@@ -0,0 +1,123 @@
+package de.kherud.llama.json;
+
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import de.kherud.llama.Pair;
+import org.junit.Test;
+
+import java.util.List;
+
+import static org.junit.Assert.*;
+
+/**
+ * Unit tests for {@link RerankResponseParser}.
+ * No JVM native library or model file needed — JSON string literals only.
+ */
+public class RerankResponseParserTest {
+
+    private static final ObjectMapper MAPPER = new ObjectMapper();
+    private final RerankResponseParser parser = new RerankResponseParser();
+
+    // ------------------------------------------------------------------
+    // parse(String)
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testParseString_singleEntry() {
+        String json = "[{\"document\":\"The quick brown fox\",\"index\":0,\"score\":0.92}]";
+        List<Pair<String, Float>> result = parser.parse(json);
+        assertEquals(1, result.size());
+        assertEquals("The quick brown fox", result.get(0).getKey());
+        assertEquals(0.92f, result.get(0).getValue(), 0.001f);
+    }
+
+    @Test
+    public void testParseString_multipleEntries() {
+        String json = "[" +
+                "{\"document\":\"First\",\"index\":0,\"score\":0.9}," +
+                "{\"document\":\"Second\",\"index\":1,\"score\":0.5}," +
+                "{\"document\":\"Third\",\"index\":2,\"score\":0.1}" +
+                "]";
+        List<Pair<String, Float>> result = parser.parse(json);
+        assertEquals(3, result.size());
+        assertEquals("First",  result.get(0).getKey());
+        assertEquals("Second", result.get(1).getKey());
+        assertEquals("Third",  result.get(2).getKey());
+        assertEquals(0.9f, result.get(0).getValue(), 0.001f);
+        assertEquals(0.5f, result.get(1).getValue(), 0.001f);
+        assertEquals(0.1f, result.get(2).getValue(), 0.001f);
+    }
+
+    @Test
+    public void testParseString_emptyArray() {
+        List<Pair<String, Float>> result = parser.parse("[]");
+        assertTrue(result.isEmpty());
+    }
+
+    @Test
+    public void testParseString_malformed() {
+        List<Pair<String, Float>> result = parser.parse("{not json");
+        assertTrue(result.isEmpty());
+    }
+
+    @Test
+    public void testParseString_notAnArray() {
+        List<Pair<String, Float>> result = parser.parse("{\"document\":\"x\",\"score\":0.5}");
+        assertTrue(result.isEmpty());
+    }
+
+    @Test
+    public void testParseString_documentWithSpecialChars() {
+        String json = "[{\"document\":\"line1\\nline2\\t\\\"quoted\\\"\",\"index\":0,\"score\":0.75}]";
+        List<Pair<String, Float>> result = parser.parse(json);
+        assertEquals(1, result.size());
+        assertEquals("line1\nline2\t\"quoted\"", result.get(0).getKey());
+    }
+
+    @Test
+    public void testParseString_scoreZero() {
+        String json = "[{\"document\":\"irrelevant\",\"index\":0,\"score\":0.0}]";
+        List<Pair<String, Float>> result = parser.parse(json);
+        assertEquals(1, result.size());
+        assertEquals(0.0f, result.get(0).getValue(), 0.001f);
+    }
+
+    // ------------------------------------------------------------------
+    // parse(JsonNode)
+    // ------------------------------------------------------------------
+
+    @Test
+    public void testParseNode_preservesOrder() throws Exception {
+        String json = "[" +
+                "{\"document\":\"A\",\"index\":0,\"score\":0.8}," +
+                "{\"document\":\"B\",\"index\":1,\"score\":0.3}" +
+                "]";
+        JsonNode arr = MAPPER.readTree(json);
+        List<Pair<String, Float>> result = parser.parse(arr);
+        assertEquals(2, result.size());
+        assertEquals("A", result.get(0).getKey());
+        assertEquals("B", result.get(1).getKey());
+    }
+
+    @Test
+    public void testParseNode_notArray() throws Exception {
+        JsonNode obj = MAPPER.readTree("{\"document\":\"x\",\"score\":0.5}");
+        assertTrue(parser.parse(obj).isEmpty());
+    }
+
+    @Test
+    public void testParseNode_missingScore_defaultsToZero() throws Exception {
+        JsonNode arr = MAPPER.readTree("[{\"document\":\"doc\",\"index\":0}]");
+        List<Pair<String, Float>> result = parser.parse(arr);
+        assertEquals(1, result.size());
+        assertEquals(0.0f, result.get(0).getValue(), 0.001f);
+    }
+
+    @Test
+    public void testParseNode_missingDocument_defaultsToEmpty() throws Exception {
+        JsonNode arr = MAPPER.readTree("[{\"index\":0,\"score\":0.5}]");
+        List<Pair<String, Float>> result = parser.parse(arr);
+        assertEquals(1, result.size());
+        assertEquals("", result.get(0).getKey());
+    }
+}