Fix Cloudflare Workers AI chat response parsing

litellm/llms/cloudflare/chat/transformation.py

@@ -71,7 +71,7 @@ class CloudflareChatConfig(BaseConfig):
             )
         headers = {
             "accept": "application/json",
-            "content-type": "apbplication/json",
+            "content-type": "application/json",
             "Authorization": "Bearer " + api_key,
         }
         return headers
@@ -149,9 +149,12 @@ class CloudflareChatConfig(BaseConfig):
     ) -> ModelResponse:
         completion_response = raw_response.json()
 
-        # Support both "response" and "response_text" keys (newer models like Nemotron use "response_text")
+        # Support Workers AI response variants:
+        # - result.response
+        # - result.response_text
+        # - result.choices[0].message.content for OpenAI-compatible chat models
         result = completion_response["result"]
-        model_response.choices[0].message.content = result.get("response") if result.get("response") is not None else result.get("response_text", "")  # type: ignore
+        model_response.choices[0].message.content = self._get_response_text(result)  # type: ignore
 
         prompt_tokens = litellm.utils.get_token_count(messages=messages, model=model)
         completion_tokens = len(
@@ -168,6 +171,28 @@ class CloudflareChatConfig(BaseConfig):
         setattr(model_response, "usage", usage)
         return model_response
 
+    @staticmethod
+    def _get_response_text(result: dict) -> str:
+        response = result.get("response")
+        if response is not None:
+            return response
+        response_text = result.get("response_text")
+        if response_text is not None:
+            return response_text
+        choices = result.get("choices")
+        if isinstance(choices, list) and len(choices) > 0:
+            first_choice = choices[0]
+            if isinstance(first_choice, dict):
+                message = first_choice.get("message")
+                if isinstance(message, dict):
+                    content = message.get("content")
+                    if content is not None:
+                        return content
+                text = first_choice.get("text")
+                if text is not None:
+                    return text
+        return ""
+
     def get_error_class(
         self, error_message: str, status_code: int, headers: Union[dict, httpx.Headers]
     ) -> BaseLLMException:

tests/llm_translation/test_cloudflare.py

@@ -35,6 +35,24 @@ def _chat_response() -> Dict[str, Any]:
     }
 
 
+def _chat_choices_response() -> Dict[str, Any]:
+    return {
+        "result": {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "content": "I am a large language model created to assist you.",
+                    }
+                }
+            ],
+        },
+        "success": True,
+        "errors": [],
+        "messages": [],
+    }
+
+
 def _streaming_chunks() -> list[str]:
     return [
         json.dumps({"response": "I am"}),
@@ -86,6 +104,41 @@ def test_completion_cloudflare(sync_mode):
     assert "language model" in response.choices[0].message.content.lower()
 
 
+@pytest.mark.parametrize("sync_mode", [True, False])
+def test_completion_cloudflare_choices_response(sync_mode):
+    messages = [{"role": "user", "content": "what llm are you"}]
+    mock_resp = _make_mock_response(_chat_choices_response())
+
+    if sync_mode:
+        with patch.object(HTTPHandler, "post", return_value=mock_resp) as mock_post:
+            response = completion(
+                model="cloudflare/@cf/openai/gpt-oss-20b",
+                messages=messages,
+                max_tokens=15,
+                api_base=FAKE_API_BASE,
+                api_key=FAKE_API_KEY,
+            )
+            mock_post.assert_called_once()
+    else:
+        with patch.object(
+            AsyncHTTPHandler, "post", new_callable=AsyncMock, return_value=mock_resp
+        ) as mock_post:
+            response = asyncio.run(
+                acompletion(
+                    model="cloudflare/@cf/openai/gpt-oss-20b",
+                    messages=messages,
+                    max_tokens=15,
+                    api_base=FAKE_API_BASE,
+                    api_key=FAKE_API_KEY,
+                )
+            )
+            mock_post.assert_called_once()
+
+    assert response is not None
+    assert response.choices[0].message.content is not None
+    assert "language model" in response.choices[0].message.content.lower()
+
+
 @pytest.mark.parametrize("sync_mode", [True, False])
 def test_completion_cloudflare_stream(sync_mode):
     messages = [{"role": "user", "content": "what llm are you"}]

tests/test_litellm/llms/cloudflare/test_cloudflare_transformation.py

@@ -25,3 +25,18 @@ def test_get_complete_url_encodes_model_path_segment():
             optional_params={},
             litellm_params={},
         )
+
+
+def test_validate_environment_sets_json_content_type():
+    config = CloudflareChatConfig()
+
+    headers = config.validate_environment(
+        headers={},
+        model="@cf/meta/llama-2-7b-chat-int8",
+        messages=[{"role": "user", "content": "ping"}],
+        optional_params={},
+        litellm_params={},
+        api_key="cf-key",
+    )
+
+    assert headers["content-type"] == "application/json"