fix(llm): register official 200k context window for o1, o1-pro, and o3 (#7303)

lib/crewai/src/crewai/llm.py

@@ -177,8 +177,11 @@ def _ensure_litellm() -> bool:
     "gpt-4.1": 1047576,  # Based on official docs
     "gpt-4.1-mini-2025-04-14": 1047576,
     "gpt-4.1-nano-2025-04-14": 1047576,
+    "o1": 200000,
+    "o1-pro": 200000,
     "o1-preview": 128000,
     "o1-mini": 128000,
+    "o3": 200000,
     "o3-mini": 200000,
     "o4-mini": 200000,
     "gemini-3-pro-preview": 1048576,

lib/crewai/src/crewai/llms/providers/azure/completion.py

@@ -1324,7 +1324,13 @@ def get_context_window_size(self) -> int:
             "gpt-4-turbo": 128000,
             "gpt-5.6": 1050000,
             "gpt-4o": 128000,
+            "o1-preview": 128000,
+            "o1-mini": 128000,
+            "o1-pro": 200000,
+            "o3-mini": 200000,
             "gpt-4": 8192,
+            "o1": 200000,
+            "o3": 200000,
         }
 
         for model_prefix, size in context_windows.items():

lib/crewai/src/crewai/llms/providers/openai/completion.py

@@ -2688,12 +2688,15 @@ def get_context_window_size(self) -> int:
             "o1-preview": 128000,
             "gpt-5.6": 1050000,
             "o1-mini": 128000,
+            "o1-pro": 200000,
             "o3-mini": 200000,
             "o4-mini": 200000,
             "gpt-4.1": 1047576,
             "gpt-4o": 128000,
             "gpt-5": 1047576,
             "gpt-4": 8192,
+            "o1": 200000,
+            "o3": 200000,
         }
 
         for model_prefix, size in context_windows.items():

lib/crewai/tests/test_llm.py

@@ -386,6 +386,36 @@ def test_unrecognized_provider_prefix_is_not_stripped() -> None:
     )
 
 
+@pytest.mark.parametrize(
+    "model",
+    ["o1", "o1-pro", "o3"],
+)
+def test_o_series_reasoning_models_context_window(model: str) -> None:
+    """Official reasoning models (o1, o1-pro, o3) use 200,000 token context window."""
+    llm = LLM(model=model, is_litellm=True)
+    assert llm.get_context_window_size() == int(200000 * CONTEXT_WINDOW_USAGE_RATIO)
+
+
+def test_o_series_does_not_override_preview_and_mini() -> None:
+    """More specific o1-preview and o1-mini must keep their 128,000 window."""
+    llm_preview = LLM(model="o1-preview", is_litellm=True)
+    assert llm_preview.get_context_window_size() == int(128000 * CONTEXT_WINDOW_USAGE_RATIO)
+
+    llm_mini = LLM(model="o1-mini", is_litellm=True)
+    assert llm_mini.get_context_window_size() == int(128000 * CONTEXT_WINDOW_USAGE_RATIO)
+
+
+@pytest.mark.parametrize(
+    "model",
+    ["openai/o1", "openai/o1-pro", "openai/o3"],
+)
+def test_openai_o_series_reasoning_models_context_window(model: str) -> None:
+    """Native OpenAI provider recognizes 200,000 token window for o1, o1-pro, and o3."""
+    llm = LLM(model=model)
+    assert llm.get_context_window_size() == int(200000 * CONTEXT_WINDOW_USAGE_RATIO)
+
+
+
 @pytest.fixture
 def get_weather_tool_schema():
     return {