feat: add reasoning parameter support for chat completions

src/together/resources/chat/completions.py

@@ -41,6 +41,7 @@ def create(
         response_format: Dict[str, Any] | None = None,
         tools: List[Dict[str, Any]] | None = None,
         tool_choice: str | Dict[str, str | Dict[str, str]] | None = None,
+        reasoning: Dict[str, Any] | None = None,
         **kwargs: Any,
     ) -> ChatCompletionResponse | Iterator[ChatCompletionChunk]:
         """
@@ -103,6 +104,9 @@ def create(
                     via {"type": "function", "function": {"name": "my_function"}} forces the model to call that function.
                     Sets to `auto` if None.
                 Defaults to None.
+            reasoning (Dict[str, Any], optional): Configuration for reasoning models. Should contain "enabled" key.
+                    Can optionally contain "effort" key with values "low", "medium", "high", or "auto".
+                Defaults to None.
 
         Returns:
             ChatCompletionResponse | Iterator[ChatCompletionChunk]: Object containing the completions
@@ -135,6 +139,7 @@ def create(
             response_format=response_format,
             tools=tools,
             tool_choice=tool_choice,
+            reasoning=reasoning,
             **kwargs,
         ).model_dump(exclude_none=True)
 
@@ -183,6 +188,7 @@ async def create(
         response_format: Dict[str, Any] | None = None,
         tools: Dict[str, str | Dict[str, str | Dict[str, Any]]] | None = None,
         tool_choice: str | Dict[str, str | Dict[str, str]] | None = None,
+        reasoning: Dict[str, Any] | None = None,
         **kwargs: Any,
     ) -> AsyncGenerator[ChatCompletionChunk, None] | ChatCompletionResponse:
         """
@@ -245,6 +251,9 @@ async def create(
                     via {"type": "function", "function": {"name": "my_function"}} forces the model to call that function.
                     Sets to `auto` if None.
                 Defaults to None.
+            reasoning (Dict[str, Any], optional): Configuration for reasoning models. Should contain "enabled" key.
+                    Can optionally contain "effort" key with values "low", "medium", "high", or "auto".
+                Defaults to None.
 
         Returns:
             AsyncGenerator[ChatCompletionChunk, None] | ChatCompletionResponse: Object containing the completions
@@ -277,6 +286,7 @@ async def create(
             response_format=response_format,
             tools=tools,
             tool_choice=tool_choice,
+            reasoning=reasoning,
             **kwargs,
         ).model_dump(exclude_none=True)
 

src/together/types/__init__.py

@@ -22,6 +22,8 @@
     ChatCompletionChunk,
     ChatCompletionRequest,
     ChatCompletionResponse,
+    Reasoning,
+    ReasoningEffort,
 )
 from together.types.common import TogetherRequest
 from together.types.completions import (
@@ -95,6 +97,8 @@
     "ChatCompletionChunk",
     "ChatCompletionRequest",
     "ChatCompletionResponse",
+    "Reasoning",
+    "ReasoningEffort",
     "EmbeddingRequest",
     "EmbeddingResponse",
     "FinetuneCheckpoint",

src/together/types/chat_completions.py

@@ -73,6 +73,7 @@ class ChatCompletionMessage(BaseModel):
     role: MessageRole
     content: str | List[ChatCompletionMessageContent] | None = None
     tool_calls: List[ToolCalls] | None = None
+    reasoning: str | None = None
 
 
 class ResponseFormat(BaseModel):
@@ -114,6 +115,18 @@ class ToolChoiceEnum(str, Enum):
     Required = "required"
 
 
+class ReasoningEffort(str, Enum):
+    LOW = "low"
+    MEDIUM = "medium"
+    HIGH = "high"
+    AUTO = "auto"
+
+
+class Reasoning(BaseModel):
+    enabled: bool
+    effort: ReasoningEffort | None = None
+
+
 class ChatCompletionRequest(BaseModel):
     # list of messages
     messages: List[ChatCompletionMessage]
@@ -148,6 +161,8 @@ class ChatCompletionRequest(BaseModel):
     response_format: ResponseFormat | None = None
     tools: List[Tools] | None = None
     tool_choice: ToolChoice | ToolChoiceEnum | None = None
+    # reasoning configuration
+    reasoning: Reasoning | None = None
 
     # Raise warning if repetition_penalty is used with presence_penalty or frequency_penalty
     @model_validator(mode="after")
@@ -183,6 +198,8 @@ class ChatCompletionResponse(BaseModel):
     prompt: List[PromptPart] | List[None] | None = None
     # token usage data
     usage: UsageData | None = None
+    # metadata (may include weight_version, reasoning stats, etc.)
+    metadata: Dict[str, Any] | None = None
 
 
 class ChatCompletionChoicesChunk(BaseModel):

src/together/types/common.py

@@ -53,6 +53,7 @@ class PromptPart(BaseModel):
 
 class DeltaContent(BaseModel):
     content: str | None = None
+    reasoning: str | None = None
 
 
 class TogetherRequest(BaseModel):