diff --git a/backend/src/adapters/upstream/openai.test.ts b/backend/src/adapters/upstream/openai.test.ts
new file mode 100644
index 0000000..ae8102f
--- /dev/null
+++ b/backend/src/adapters/upstream/openai.test.ts
@@ -0,0 +1,98 @@
+import { describe, expect, test } from "bun:test";
+import { openaiUpstreamAdapter } from "./openai";
+
+describe("openaiUpstreamAdapter reasoning compatibility", () => {
+  test("parses non-stream reasoning field into thinking block", async () => {
+    const response = new Response(
+      JSON.stringify({
+        id: "chatcmpl-1",
+        object: "chat.completion",
+        created: 1700000000,
+        model: "test-model",
+        choices: [
+          {
+            index: 0,
+            message: {
+              role: "assistant",
+              content: "final answer",
+              reasoning: "chain of thought summary",
+            },
+            finish_reason: "stop",
+          },
+        ],
+        usage: {
+          prompt_tokens: 1,
+          completion_tokens: 2,
+          total_tokens: 3,
+        },
+      }),
+    );
+
+    const parsed = await openaiUpstreamAdapter.parseResponse(response);
+    expect(parsed.content).toEqual([
+      { type: "thinking", thinking: "chain of thought summary" },
+      { type: "text", text: "final answer" },
+    ]);
+  });
+
+  test("prefers reasoning_content over reasoning when both exist", async () => {
+    const response = new Response(
+      JSON.stringify({
+        id: "chatcmpl-2",
+        object: "chat.completion",
+        created: 1700000001,
+        model: "test-model",
+        choices: [
+          {
+            index: 0,
+            message: {
+              role: "assistant",
+              content: "final answer",
+              reasoning_content: "preferred reasoning content",
+              reasoning: "fallback reasoning",
+            },
+            finish_reason: "stop",
+          },
+        ],
+        usage: {
+          prompt_tokens: 1,
+          completion_tokens: 2,
+          total_tokens: 3,
+        },
+      }),
+    );
+
+    const parsed = await openaiUpstreamAdapter.parseResponse(response);
+    expect(parsed.content).toEqual([
+      { type: "thinking", thinking: "preferred reasoning content" },
+      { type: "text", text: "final answer" },
+    ]);
+  });
+
+  test("parses stream delta reasoning field into thinking_delta", async () => {
+    const stream = [
+      'data: {"id":"chatcmpl-3","object":"chat.completion.chunk","created":1700000002,"model":"test-model","choices":[{"index":0,"delta":{"role":"assistant","reasoning":"stream reasoning"},"finish_reason":null}]}',
+      'data: {"id":"chatcmpl-3","object":"chat.completion.chunk","created":1700000002,"model":"test-model","choices":[{"index":0,"delta":{"content":"stream text"},"finish_reason":"stop"}]}',
+      "data: [DONE]",
+    ].join("\n");
+
+    const response = new Response(stream);
+    const chunks: Array<unknown> = [];
+    for await (const chunk of openaiUpstreamAdapter.parseStreamResponse(
+      response,
+    )) {
+      chunks.push(chunk);
+    }
+
+    expect(chunks).toContainEqual({
+      type: "content_block_delta",
+      index: 0,
+      delta: { type: "thinking_delta", thinking: "stream reasoning" },
+    });
+    expect(chunks).toContainEqual({
+      type: "content_block_delta",
+      index: 0,
+      delta: { type: "text_delta", text: "stream text" },
+    });
+  });
+});
diff --git a/backend/src/adapters/upstream/openai.ts b/backend/src/adapters/upstream/openai.ts
index 64c935c..bfb8693 100644
--- a/backend/src/adapters/upstream/openai.ts
+++ b/backend/src/adapters/upstream/openai.ts
@@ -93,6 +93,7 @@ interface OpenAIChoice {
     content: string | null;
     tool_calls?: OpenAIToolCall[];
     reasoning_content?: string;
+    reasoning?: string;
   };
   finish_reason: string | null;
 }
@@ -117,6 +118,7 @@ interface OpenAIStreamChoice {
     content?: string | null;
     tool_calls?: OpenAIToolCallDelta[];
     reasoning_content?: string;
+    reasoning?: string;
   };
   finish_reason: string | null;
 }
@@ -291,6 +293,24 @@ function convertFinishReason(finishReason: string | null): StopReason {
   }
 }
 
+function extractReasoningText(
+  payload?: {
+    reasoning_content?: string;
+    reasoning?: string;
+  },
+): string | undefined {
+  if (!payload) {
+    return undefined;
+  }
+  if (payload.reasoning_content && payload.reasoning_content.length > 0) {
+    return payload.reasoning_content;
+  }
+  if (payload.reasoning && payload.reasoning.length > 0) {
+    return payload.reasoning;
+  }
+  return undefined;
+}
+
 /**
  * Convert OpenAI response to internal format
  */
@@ -298,11 +318,12 @@ function convertResponse(resp: OpenAIChatResponse): InternalResponse {
   const choice = resp.choices[0];
   const content: InternalContentBlock[] = [];
 
-  // Handle reasoning content (for o1/deepseek models)
-  if (choice?.message.reasoning_content) {
+  // Handle reasoning content (reasoning_content or reasoning)
+  const reasoningText = extractReasoningText(choice?.message);
+  if (reasoningText) {
     content.push({
       type: "thinking",
-      thinking: choice.message.reasoning_content,
+      thinking: reasoningText,
     } as ThinkingContentBlock);
   }
 
@@ -524,13 +545,14 @@ export const openaiUpstreamAdapter: UpstreamAdapter = {
       }
 
       // Handle reasoning content (thinking)
-      if (choice.delta.reasoning_content) {
+      const reasoningDelta = extractReasoningText(choice.delta);
+      if (reasoningDelta) {
         yield {
           type: "content_block_delta",
           index: blockIndex,
           delta: {
             type: "thinking_delta",
-            thinking: choice.delta.reasoning_content,
+            thinking: reasoningDelta,
           },
         };
       }