fix: preserve provider on LiteLLM-routed models

LiteLLM construction computed the real provider in `__new__` but never
passed it into init, so `BaseLLM` silently defaulted every shared-path
model to `openai`. Infer the provider from a `provider/model` prefix
when none is supplied so groq, cohere, mistral, and the rest report
themselves correctly to callers like the policy engine.

lib/crewai/src/crewai/llms/base_llm.py

@@ -274,7 +274,10 @@ class BaseLLM(BaseModel, ABC):
             data["stop"] = list(stop)
 
         if not data.get("provider"):
-            data["provider"] = "openai"
+            model = data.get("model") or ""
+            data["provider"] = (
+                cls._extract_provider(model) if isinstance(model, str) else "openai"
+            )
 
         known_fields = set(cls.model_fields.keys())
         extras = {k: v for k, v in data.items() if k not in known_fields}

lib/crewai/tests/test_llm.py

@@ -860,16 +860,19 @@ def test_prefixed_models_with_invalid_constants_use_litellm():
     llm = LLM(model="openai/gemini-2.5-flash", is_litellm=False)
     assert llm.is_litellm is True
     assert llm.model == "openai/gemini-2.5-flash"
+    assert llm.provider == "openai"
 
     # Test openai/ prefix with model that doesn't match patterns (e.g. no gpt- prefix) → LiteLLM
     llm2 = LLM(model="openai/custom-finetune-model", is_litellm=False)
     assert llm2.is_litellm is True
     assert llm2.model == "openai/custom-finetune-model"
+    assert llm2.provider == "openai"
 
     # Test anthropic/ prefix with non-Anthropic model → LiteLLM
     llm3 = LLM(model="anthropic/gpt-4o", is_litellm=False)
     assert llm3.is_litellm is True
     assert llm3.model == "anthropic/gpt-4o"
+    assert llm3.provider == "anthropic"
 
 
 def test_prefixed_models_with_valid_patterns_use_native_sdk():
@@ -893,11 +896,32 @@ def test_prefixed_models_with_non_native_providers_use_litellm():
     llm = LLM(model="groq/llama-3.3-70b", is_litellm=False)
     assert llm.is_litellm is True
     assert llm.model == "groq/llama-3.3-70b"
+    assert llm.provider == "groq"
 
     # Test together/ prefix (not a native provider) → LiteLLM
     llm2 = LLM(model="together/qwen-2.5-72b", is_litellm=False)
     assert llm2.is_litellm is True
     assert llm2.model == "together/qwen-2.5-72b"
+    assert llm2.provider == "together"
+
+
+@pytest.mark.parametrize(
+    ("model", "expected_provider"),
+    [
+        ("groq/llama-3.3-70b", "groq"),
+        ("cohere/command-r", "cohere"),
+        ("sambanova/Meta-Llama-3.1-70B-Instruct", "sambanova"),
+        ("mistral/mistral-large", "mistral"),
+        ("vertex_ai/gemini-1.5-pro", "vertex_ai"),
+        ("openai/custom-finetune-model", "openai"),
+        ("anthropic/gpt-4o", "anthropic"),
+    ],
+)
+def test_litellm_path_preserves_provider_from_model_prefix(model, expected_provider):
+    llm = LLM(model=model, is_litellm=False)
+    assert llm.is_litellm is True
+    assert llm.provider == expected_provider
+    assert llm.model == model
 
 
 def test_unprefixed_models_use_native_sdk():