Accept xhigh reasoning_effort and add grok-4.6 to ChatModel

CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## [Unreleased]
 ### Added
+- **`xhigh` Reasoning Effort**: Added `"xhigh"` as an accepted `reasoning_effort` value (maps to `EFFORT_XHIGH`; supported by models such as `grok-4.6`)
+- Added `grok-4.6` to the `ChatModel` known-model type literal
 - **Imagine File Storage**: Image and video generation (sync and async) now accept a `storage_options` parameter to persist generated assets to the Files API. It takes a dict with a required `filename` and optional `expires_after` (an `int` in seconds or a `datetime.timedelta`) and `public_url` (`True` to create a public URL with default expiry, or `{"expires_after": <seconds>}` for an independent URL expiry). Image and video responses expose new `file_output`, `storage_error`, `public_url`, and `public_url_error` properties.
 - **File-ID Inputs for Generation**: Image and video generation now accept Files API `file_id` references as inputs alongside URLs/base64 — `image_file_id` / `image_file_ids` for `image.sample()` / `image.sample_batch()`, and `image_file_id` / `video_file_id` / `reference_image_file_ids` for `video.generate()` / `video.extend()` (and the batch `prepare` helpers). URL and file-ID lists may be mixed in the same multi-image request (file IDs are sent first).
 - **Public File URLs**: Added `client.files.create_public_url()` and `client.files.revoke_public_url()` (sync and async) to create and revoke publicly shareable, unauthenticated URLs for stored files. `create_public_url()` accepts an optional `expires_after` (an `int` in seconds or a `datetime.timedelta`).

src/xai_sdk/chat.py

@@ -141,8 +141,10 @@ def create(
             presence_penalty: Positive values penalize new tokens based on whether they appear in the text so far,
                 increasing the model's likelihood to talk about new topics.
             reasoning_effort: Constrains how hard a reasoning model thinks before responding. Possible values are
-                `none` (no reasoning), `low` (uses fewer reasoning tokens), `medium` (default), and `high`
-                (uses more reasoning tokens). Defaults to `medium`.
+                `none` (no reasoning), `low` (uses fewer reasoning tokens), `medium`, `high`
+                (uses more reasoning tokens), and `xhigh` (maximum reasoning depth; only supported by some
+                models, e.g. `grok-4.6`). The default varies by model (e.g. `grok-4.5` and `grok-4.6`
+                default to `high`).
             search_parameters: The parameters that control search behavior.
                 This includes settings like search mode, date range, sources (e.g., web, news, or X), and whether
                 to return citations. See `SearchParameters` for detailed configuration options.
@@ -921,6 +923,8 @@ def _reasoning_effort_to_proto(effort: ReasoningEffort) -> chat_pb2.ReasoningEff
             return chat_pb2.ReasoningEffort.EFFORT_MEDIUM
         case "high":
             return chat_pb2.ReasoningEffort.EFFORT_HIGH
+        case "xhigh":
+            return chat_pb2.ReasoningEffort.EFFORT_XHIGH
         case _:
             raise ValueError(f"Invalid reasoning effort: {effort}. Must be one of: {ReasoningEffort.__args__}")
 

src/xai_sdk/proto/v5/chat_pb2.pyi

@@ -42,6 +42,7 @@ class ReasoningEffort(int, metaclass=_enum_type_wrapper.EnumTypeWrapper):
     EFFORT_MEDIUM: _ClassVar[ReasoningEffort]
     EFFORT_HIGH: _ClassVar[ReasoningEffort]
     EFFORT_NONE: _ClassVar[ReasoningEffort]
+    EFFORT_XHIGH: _ClassVar[ReasoningEffort]
 
 class AgentCount(int, metaclass=_enum_type_wrapper.EnumTypeWrapper):
     __slots__ = ()
@@ -109,6 +110,7 @@ EFFORT_LOW: ReasoningEffort
 EFFORT_MEDIUM: ReasoningEffort
 EFFORT_HIGH: ReasoningEffort
 EFFORT_NONE: ReasoningEffort
+EFFORT_XHIGH: ReasoningEffort
 AGENT_COUNT_UNSPECIFIED: AgentCount
 AGENT_COUNT_4: AgentCount
 AGENT_COUNT_16: AgentCount

src/xai_sdk/proto/v6/chat_pb2.pyi

@@ -43,6 +43,7 @@ class ReasoningEffort(int, metaclass=_enum_type_wrapper.EnumTypeWrapper):
     EFFORT_MEDIUM: _ClassVar[ReasoningEffort]
     EFFORT_HIGH: _ClassVar[ReasoningEffort]
     EFFORT_NONE: _ClassVar[ReasoningEffort]
+    EFFORT_XHIGH: _ClassVar[ReasoningEffort]
 
 class AgentCount(int, metaclass=_enum_type_wrapper.EnumTypeWrapper):
     __slots__ = ()
@@ -110,6 +111,7 @@ EFFORT_LOW: ReasoningEffort
 EFFORT_MEDIUM: ReasoningEffort
 EFFORT_HIGH: ReasoningEffort
 EFFORT_NONE: ReasoningEffort
+EFFORT_XHIGH: ReasoningEffort
 AGENT_COUNT_UNSPECIFIED: AgentCount
 AGENT_COUNT_4: AgentCount
 AGENT_COUNT_16: AgentCount

src/xai_sdk/types/chat.py

@@ -15,7 +15,7 @@
 ]
 
 AgentCount: TypeAlias = Literal[4, 16]
-ReasoningEffort: TypeAlias = Literal["none", "low", "medium", "high"]
+ReasoningEffort: TypeAlias = Literal["none", "low", "medium", "high", "xhigh"]
 ImageDetail: TypeAlias = Literal["auto", "low", "high"]
 Content: TypeAlias = Union[str, chat_pb2.Content]
 ToolMode: TypeAlias = Literal["auto", "none", "required"]

src/xai_sdk/types/model.py

@@ -30,6 +30,7 @@
     "grok-4.3-latest",
     "grok-4.5",
     "grok-4.5-latest",
+    "grok-4.6",
     "grok-code-fast-1",
     "grok-build-0.1",
     "grok-3",

tests/aio/chat_test.py

@@ -1362,10 +1362,12 @@ async def test_multi_turn_conversation_creates_multiple_spans_with_same_conversa
         "low",
         "medium",
         "high",
+        "xhigh",
         chat_pb2.ReasoningEffort.EFFORT_NONE,
         chat_pb2.ReasoningEffort.EFFORT_LOW,
         chat_pb2.ReasoningEffort.EFFORT_MEDIUM,
         chat_pb2.ReasoningEffort.EFFORT_HIGH,
+        chat_pb2.ReasoningEffort.EFFORT_XHIGH,
     ],
 )
 def test_chat_create_with_reasoning(
@@ -1385,6 +1387,8 @@ def test_chat_create_with_reasoning(
         assert chat_completion_request.reasoning_effort == chat_pb2.ReasoningEffort.EFFORT_MEDIUM
     elif reasoning_effort == "high":
         assert chat_completion_request.reasoning_effort == chat_pb2.ReasoningEffort.EFFORT_HIGH
+    elif reasoning_effort == "xhigh":
+        assert chat_completion_request.reasoning_effort == chat_pb2.ReasoningEffort.EFFORT_XHIGH
     else:
         assert chat_completion_request.reasoning_effort == reasoning_effort
 
@@ -1395,7 +1399,9 @@ def test_chat_with_reasoning_invalid_value(client: AsyncClient):
             "grok-4.3",
             reasoning_effort="invalid",  # type: ignore
         )
-    assert str(e.value) == "Invalid reasoning effort: invalid. Must be one of: ('none', 'low', 'medium', 'high')"
+    assert (
+        str(e.value) == "Invalid reasoning effort: invalid. Must be one of: ('none', 'low', 'medium', 'high', 'xhigh')"
+    )
 
 
 def test_chat_create_with_tools(client: AsyncClient):

tests/sync/chat_test.py

@@ -1321,10 +1321,12 @@ def test_multi_turn_conversation_creates_multiple_spans_with_same_conversation_i
         "low",
         "medium",
         "high",
+        "xhigh",
         chat_pb2.ReasoningEffort.EFFORT_NONE,
         chat_pb2.ReasoningEffort.EFFORT_LOW,
         chat_pb2.ReasoningEffort.EFFORT_MEDIUM,
         chat_pb2.ReasoningEffort.EFFORT_HIGH,
+        chat_pb2.ReasoningEffort.EFFORT_XHIGH,
     ],
 )
 def test_chat_create_with_reasoning(
@@ -1344,6 +1346,8 @@ def test_chat_create_with_reasoning(
         assert chat_completion_request.reasoning_effort == chat_pb2.ReasoningEffort.EFFORT_MEDIUM
     elif reasoning_effort == "high":
         assert chat_completion_request.reasoning_effort == chat_pb2.ReasoningEffort.EFFORT_HIGH
+    elif reasoning_effort == "xhigh":
+        assert chat_completion_request.reasoning_effort == chat_pb2.ReasoningEffort.EFFORT_XHIGH
     else:
         assert chat_completion_request.reasoning_effort == reasoning_effort
 
@@ -1354,7 +1358,9 @@ def test_chat_with_reasoning_invalid_value(client: Client):
             "grok-4.3",
             reasoning_effort="invalid",  # type: ignore
         )
-    assert str(e.value) == "Invalid reasoning effort: invalid. Must be one of: ('none', 'low', 'medium', 'high')"
+    assert (
+        str(e.value) == "Invalid reasoning effort: invalid. Must be one of: ('none', 'low', 'medium', 'high', 'xhigh')"
+    )
 
 
 def test_chat_create_with_tools(client: Client):