feat(cohere): support reasoning

src/any_llm/providers/cohere/cohere.py

@@ -10,6 +10,7 @@
 MISSING_PACKAGES_ERROR = None
 try:
     import cohere
+    from cohere import V2ChatResponse
 
     from .utils import (
         _convert_models_list,
@@ -37,7 +38,7 @@ class CohereProvider(AnyLLM):
     SUPPORTS_COMPLETION_STREAMING = True
     SUPPORTS_COMPLETION = True
     SUPPORTS_RESPONSES = False
-    SUPPORTS_COMPLETION_REASONING = False
+    SUPPORTS_COMPLETION_REASONING = True
     SUPPORTS_COMPLETION_IMAGE = False
     SUPPORTS_COMPLETION_PDF = False
     SUPPORTS_EMBEDDING = False
@@ -60,10 +61,10 @@ def _convert_completion_params(params: CompletionParams, **kwargs: Any) -> dict[
         return converted_params
 
     @staticmethod
-    def _convert_completion_response(response: Any, **kwargs: Any) -> ChatCompletion:
+    def _convert_completion_response(response: V2ChatResponse, **kwargs: Any) -> ChatCompletion:
         """Convert Cohere response to OpenAI format."""
         # We need the model parameter for conversion
-        model = kwargs.get("model", getattr(response, "model", "cohere-model"))
+        model = kwargs.get("model", "unknown")
         return _convert_response(response, model)
 
     @staticmethod

src/any_llm/providers/cohere/utils.py

@@ -1,6 +1,7 @@
 from collections.abc import Sequence
 from typing import Any
 
+from cohere import V2ChatResponse
 from cohere.types import ListModelsResponse as CohereListModelsResponse
 
 from any_llm.types.completion import (
@@ -11,6 +12,7 @@
     Choice,
     CompletionUsage,
     Function,
+    Reasoning,
 )
 from any_llm.types.model import Model
 
@@ -61,9 +63,12 @@ def _create_openai_chunk_from_cohere_chunk(chunk: Any) -> ChatCompletionChunk:
             and chunk.delta.message
             and hasattr(chunk.delta.message, "content")
             and chunk.delta.message.content
-            and hasattr(chunk.delta.message.content, "text")
         ):
-            delta["content"] = chunk.delta.message.content.text
+            content_obj = chunk.delta.message.content
+            if hasattr(content_obj, "text") and content_obj.text:
+                delta["content"] = content_obj.text
+            elif hasattr(content_obj, "thinking") and content_obj.thinking:
+                delta["reasoning"] = {"content": content_obj.thinking}
 
     elif chunk_type == "tool-call-start":
         if (
@@ -144,7 +149,7 @@ def _create_openai_chunk_from_cohere_chunk(chunk: Any) -> ChatCompletionChunk:
     return ChatCompletionChunk.model_validate(chunk_dict)
 
 
-def _convert_response(response: Any, model: str) -> ChatCompletion:
+def _convert_response(response: V2ChatResponse, model: str) -> ChatCompletion:
     """Convert Cohere response to OpenAI ChatCompletion format directly."""
     prompt_tokens = 0
     completion_tokens = 0
@@ -166,11 +171,13 @@ def _convert_response(response: Any, model: str) -> ChatCompletion:
             content=response.message.tool_plan,
             tool_calls=[
                 ChatCompletionMessageFunctionToolCall(
-                    id=tool_call.id,
+                    id=tool_call.id or "",
                     type="function",
                     function=Function(
-                        name=tool_call.function.name if tool_call.function else "",
-                        arguments=tool_call.function.arguments if tool_call.function else "",
+                        name=tool_call.function.name if tool_call.function and tool_call.function.name else "",
+                        arguments=tool_call.function.arguments
+                        if tool_call.function and tool_call.function.arguments
+                        else "",
                     ),
                 )
             ],
@@ -185,10 +192,17 @@ def _convert_response(response: Any, model: str) -> ChatCompletion:
             usage=usage,
         )
     content = ""
-    if response.message.content and len(response.message.content) > 0:
-        content = response.message.content[0].text
+    reasoning_content = None
 
-    message = ChatCompletionMessage(role="assistant", content=content, tool_calls=None)
+    if response.message.content and len(response.message.content) > 0:
+        for item in response.message.content:
+            if hasattr(item, "type"):
+                if item.type == "text" and hasattr(item, "text"):
+                    content += item.text
+                elif item.type == "thinking" and hasattr(item, "thinking"):
+                    reasoning_content = Reasoning(content=item.thinking)
+
+    message = ChatCompletionMessage(role="assistant", content=content, tool_calls=None, reasoning=reasoning_content)
     choice = Choice(
         index=0,
         finish_reason="stop",

tests/conftest.py

@@ -24,6 +24,7 @@ def provider_reasoning_model_map() -> dict[LLMProvider, str]:
         LLMProvider.LLAMACPP: "N/A",
         LLMProvider.LMSTUDIO: "openai/gpt-oss-20b",  # You must have LM Studio running and the server enabled
         LLMProvider.AZUREOPENAI: "azure/<your_deployment_name>",
+        LLMProvider.COHERE: "command-a-reasoning-08-2025",
     }