diff --git a/README.md b/README.md
index 3c259f9..6373aa6 100644
--- a/README.md
+++ b/README.md
@@ -201,7 +201,7 @@ export RECAPTION_PROVIDER="minimax"
 export MINIMAX_API_KEY="your-minimax-api-key"
 ```
 
-Uses the [MiniMax](https://www.minimax.io/) OpenAI-compatible API with the `MiniMax-M2.7` model by default. You can also use `MiniMax-M2.7-highspeed` for faster responses.
+Uses the [MiniMax](https://www.minimax.io/) OpenAI-compatible API with the `MiniMax-M3` model by default. You can also use `MiniMax-M2.7` or `MiniMax-M2.7-highspeed` for the previous generation (set via `OPENAI_COMPATIBLE_MODEL`).
 
 **Option 3: Any OpenAI-compatible API**
 
diff --git a/agent/config.py b/agent/config.py
index af5ca64..67d8f9c 100644
--- a/agent/config.py
+++ b/agent/config.py
@@ -13,7 +13,7 @@
 # ---------------------------------------------------------------------------
 # Supported values: "gemini" (default), "openai_compatible", "minimax"
 # When set to "minimax", uses the OpenAI-compatible provider with MiniMax
-# defaults (base URL: https://api.minimax.io/v1, model: MiniMax-M2.7).
+# defaults (base URL: https://api.minimax.io/v1, model: MiniMax-M3).
 RECAPTION_PROVIDER: str = os.environ.get("RECAPTION_PROVIDER", "gemini")
 
 # ---------------------------------------------------------------------------
diff --git a/agent/llm_provider.py b/agent/llm_provider.py
index 3af99fa..5245c4a 100644
--- a/agent/llm_provider.py
+++ b/agent/llm_provider.py
@@ -119,7 +119,7 @@ class OpenAICompatibleProvider(LLMProvider):
     - ``OPENAI_COMPATIBLE_API_KEY`` or ``MINIMAX_API_KEY``: API key.
     - ``OPENAI_COMPATIBLE_BASE_URL``: Base URL (default: MiniMax
       ``https://api.minimax.io/v1``).
-    - ``OPENAI_COMPATIBLE_MODEL``: Model name (default: ``MiniMax-M2.7``).
+    - ``OPENAI_COMPATIBLE_MODEL``: Model name (default: ``MiniMax-M3``).
     """
 
     def __init__(
@@ -141,7 +141,7 @@ def __init__(
         )
         self._model = (
             model
-            or os.environ.get("OPENAI_COMPATIBLE_MODEL", "MiniMax-M2.7")
+            or os.environ.get("OPENAI_COMPATIBLE_MODEL", "MiniMax-M3")
         )
 
     def _get_client(self) -> Any:
@@ -202,7 +202,7 @@ def get_recaption_provider() -> LLMProvider:
 
     When ``"minimax"`` is selected, the OpenAI-compatible provider is used
     with MiniMax defaults (``https://api.minimax.io/v1``, model
-    ``MiniMax-M2.7``).
+    ``MiniMax-M3``).
     """
     provider_name = os.environ.get("RECAPTION_PROVIDER", "gemini").lower()
     provider_cls = _PROVIDERS.get(provider_name)
diff --git a/tests/test_integration.py b/tests/test_integration.py
index 714643f..fd37c00 100644
--- a/tests/test_integration.py
+++ b/tests/test_integration.py
@@ -41,7 +41,7 @@ def test_minimax_provider_generate(self):
         provider = OpenAICompatibleProvider(
             api_key=os.environ["MINIMAX_API_KEY"],
             base_url="https://api.minimax.io/v1",
-            model="MiniMax-M2.7",
+            model="MiniMax-M3",
         )
         result = provider.generate(
             system_prompt="You are a helpful assistant.",
@@ -79,7 +79,7 @@ def test_minimax_temperature_boundaries(self):
         provider = OpenAICompatibleProvider(
             api_key=os.environ["MINIMAX_API_KEY"],
             base_url="https://api.minimax.io/v1",
-            model="MiniMax-M2.7",
+            model="MiniMax-M3",
         )
         # Should not raise with edge temperature values
         result = provider.generate(
diff --git a/tests/test_llm_provider.py b/tests/test_llm_provider.py
index 2f248f7..d1c66c5 100644
--- a/tests/test_llm_provider.py
+++ b/tests/test_llm_provider.py
@@ -160,7 +160,7 @@ def test_defaults_to_minimax(self):
             provider = OpenAICompatibleProvider()
             assert provider._api_key == "mm-key"
             assert "minimax" in provider._base_url.lower()
-            assert provider._model == "MiniMax-M2.7"
+            assert provider._model == "MiniMax-M3"
 
     def test_explicit_params(self):
         provider = OpenAICompatibleProvider(