pydantic-ai-slim 1.7.0__py3-none-any.whl → 1.11.0__py3-none-any.whl

pydantic_ai/durable_exec/temporal/_function_toolset.py

@@ -1,57 +1,22 @@
 from __future__ import annotations
 
 from collections.abc import Callable
-from dataclasses import dataclass
-from typing import Annotated, Any, Literal
+from typing import Any, Literal
 
-from pydantic import ConfigDict, Discriminator, with_config
 from temporalio import activity, workflow
 from temporalio.workflow import ActivityConfig
-from typing_extensions import assert_never
 
 from pydantic_ai import FunctionToolset, ToolsetTool
-from pydantic_ai.exceptions import ApprovalRequired, CallDeferred, ModelRetry, UserError
+from pydantic_ai.exceptions import UserError
 from pydantic_ai.tools import AgentDepsT, RunContext
 from pydantic_ai.toolsets.function import FunctionToolsetTool
 
 from ._run_context import TemporalRunContext
-from ._toolset import TemporalWrapperToolset
-
-
-@dataclass
-@with_config(ConfigDict(arbitrary_types_allowed=True))
-class _CallToolParams:
-    name: str
-    tool_args: dict[str, Any]
-    serialized_run_context: Any
-
-
-@dataclass
-class _ApprovalRequired:
-    kind: Literal['approval_required'] = 'approval_required'
-
-
-@dataclass
-class _CallDeferred:
-    kind: Literal['call_deferred'] = 'call_deferred'
-
-
-@dataclass
-class _ModelRetry:
-    message: str
-    kind: Literal['model_retry'] = 'model_retry'
-
-
-@dataclass
-class _ToolReturn:
-    result: Any
-    kind: Literal['tool_return'] = 'tool_return'
-
-
-_CallToolResult = Annotated[
-    _ApprovalRequired | _CallDeferred | _ModelRetry | _ToolReturn,
-    Discriminator('kind'),
-]
+from ._toolset import (
+    CallToolParams,
+    CallToolResult,
+    TemporalWrapperToolset,
+)
 
 
 class TemporalFunctionToolset(TemporalWrapperToolset[AgentDepsT]):
@@ -70,7 +35,7 @@ class TemporalFunctionToolset(TemporalWrapperToolset[AgentDepsT]):
         self.tool_activity_config = tool_activity_config
         self.run_context_type = run_context_type
 
-        async def call_tool_activity(params: _CallToolParams, deps: AgentDepsT) -> _CallToolResult:
+        async def call_tool_activity(params: CallToolParams, deps: AgentDepsT) -> CallToolResult:
             name = params.name
             ctx = self.run_context_type.deserialize_run_context(params.serialized_run_context, deps=deps)
             try:
@@ -84,15 +49,7 @@ class TemporalFunctionToolset(TemporalWrapperToolset[AgentDepsT]):
             # The tool args will already have been validated into their proper types in the `ToolManager`,
             # but `execute_activity` would have turned them into simple Python types again, so we need to re-validate them.
             args_dict = tool.args_validator.validate_python(params.tool_args)
-            try:
-                result = await self.wrapped.call_tool(name, args_dict, ctx, tool)
-                return _ToolReturn(result=result)
-            except ApprovalRequired:
-                return _ApprovalRequired()
-            except CallDeferred:
-                return _CallDeferred()
-            except ModelRetry as e:
-                return _ModelRetry(message=e.message)
+            return await self._wrap_call_tool_result(self.wrapped.call_tool(name, args_dict, ctx, tool))
 
         # Set type hint explicitly so that Temporal can take care of serialization and deserialization
         call_tool_activity.__annotations__['deps'] = deps_type
@@ -123,25 +80,18 @@ class TemporalFunctionToolset(TemporalWrapperToolset[AgentDepsT]):
 
         tool_activity_config = self.activity_config | tool_activity_config
         serialized_run_context = self.run_context_type.serialize_run_context(ctx)
-        result = await workflow.execute_activity(  # pyright: ignore[reportUnknownMemberType]
-            activity=self.call_tool_activity,
-            args=[
-                _CallToolParams(
-                    name=name,
-                    tool_args=tool_args,
-                    serialized_run_context=serialized_run_context,
-                ),
-                ctx.deps,
-            ],
-            **tool_activity_config,
+        return self._unwrap_call_tool_result(
+            await workflow.execute_activity(  # pyright: ignore[reportUnknownMemberType]
+                activity=self.call_tool_activity,
+                args=[
+                    CallToolParams(
+                        name=name,
+                        tool_args=tool_args,
+                        serialized_run_context=serialized_run_context,
+                        tool_def=None,
+                    ),
+                    ctx.deps,
+                ],
+                **tool_activity_config,
+            )
         )
-        if isinstance(result, _ApprovalRequired):
-            raise ApprovalRequired()
-        elif isinstance(result, _CallDeferred):
-            raise CallDeferred()
-        elif isinstance(result, _ModelRetry):
-            raise ModelRetry(result.message)
-        elif isinstance(result, _ToolReturn):
-            return result.result
-        else:
-            assert_never(result)

pydantic_ai/durable_exec/temporal/_mcp_server.py

@@ -11,11 +11,15 @@ from typing_extensions import Self
 
 from pydantic_ai import ToolsetTool
 from pydantic_ai.exceptions import UserError
-from pydantic_ai.mcp import MCPServer, ToolResult
+from pydantic_ai.mcp import MCPServer
 from pydantic_ai.tools import AgentDepsT, RunContext, ToolDefinition
 
 from ._run_context import TemporalRunContext
-from ._toolset import TemporalWrapperToolset
+from ._toolset import (
+    CallToolParams,
+    CallToolResult,
+    TemporalWrapperToolset,
+)
 
 
 @dataclass
@@ -24,15 +28,6 @@ class _GetToolsParams:
     serialized_run_context: Any
 
 
-@dataclass
-@with_config(ConfigDict(arbitrary_types_allowed=True))
-class _CallToolParams:
-    name: str
-    tool_args: dict[str, Any]
-    serialized_run_context: Any
-    tool_def: ToolDefinition
-
-
 class TemporalMCPServer(TemporalWrapperToolset[AgentDepsT]):
     def __init__(
         self,
@@ -72,13 +67,16 @@ class TemporalMCPServer(TemporalWrapperToolset[AgentDepsT]):
             get_tools_activity
         )
 
-        async def call_tool_activity(params: _CallToolParams, deps: AgentDepsT) -> ToolResult:
+        async def call_tool_activity(params: CallToolParams, deps: AgentDepsT) -> CallToolResult:
             run_context = self.run_context_type.deserialize_run_context(params.serialized_run_context, deps=deps)
-            return await self.wrapped.call_tool(
-                params.name,
-                params.tool_args,
-                run_context,
-                self.tool_for_tool_def(params.tool_def),
+            assert isinstance(params.tool_def, ToolDefinition)
+            return await self._wrap_call_tool_result(
+                self.wrapped.call_tool(
+                    params.name,
+                    params.tool_args,
+                    run_context,
+                    self.tool_for_tool_def(params.tool_def),
+                )
             )
 
         # Set type hint explicitly so that Temporal can take care of serialization and deserialization
@@ -125,22 +123,24 @@ class TemporalMCPServer(TemporalWrapperToolset[AgentDepsT]):
         tool_args: dict[str, Any],
         ctx: RunContext[AgentDepsT],
         tool: ToolsetTool[AgentDepsT],
-    ) -> ToolResult:
+    ) -> CallToolResult:
         if not workflow.in_workflow():
             return await super().call_tool(name, tool_args, ctx, tool)
 
         tool_activity_config = self.activity_config | self.tool_activity_config.get(name, {})
         serialized_run_context = self.run_context_type.serialize_run_context(ctx)
-        return await workflow.execute_activity(  # pyright: ignore[reportUnknownMemberType]
-            activity=self.call_tool_activity,
-            args=[
-                _CallToolParams(
-                    name=name,
-                    tool_args=tool_args,
-                    serialized_run_context=serialized_run_context,
-                    tool_def=tool.tool_def,
-                ),
-                ctx.deps,
-            ],
-            **tool_activity_config,
+        return self._unwrap_call_tool_result(
+            await workflow.execute_activity(  # pyright: ignore[reportUnknownMemberType]
+                activity=self.call_tool_activity,
+                args=[
+                    CallToolParams(
+                        name=name,
+                        tool_args=tool_args,
+                        serialized_run_context=serialized_run_context,
+                        tool_def=tool.tool_def,
+                    ),
+                    ctx.deps,
+                ],
+                **tool_activity_config,
+            )
         )

pydantic_ai/durable_exec/temporal/_run_context.py

@@ -2,14 +2,19 @@ from __future__ import annotations
 
 from typing import Any
 
+from typing_extensions import TypeVar
+
 from pydantic_ai.exceptions import UserError
-from pydantic_ai.tools import AgentDepsT, RunContext
+from pydantic_ai.tools import RunContext
+
+AgentDepsT = TypeVar('AgentDepsT', default=None, covariant=True)
+"""Type variable for the agent dependencies in `RunContext`."""
 
 
 class TemporalRunContext(RunContext[AgentDepsT]):
     """The [`RunContext`][pydantic_ai.tools.RunContext] subclass to use to serialize and deserialize the run context for use inside a Temporal activity.
 
-    By default, only the `deps`, `retries`, `tool_call_id`, `tool_name`, `tool_call_approved`, `retry`, `max_retries` and `run_step` attributes will be available.
+    By default, only the `deps`, `retries`, `tool_call_id`, `tool_name`, `tool_call_approved`, `retry`, `max_retries`, `run_step` and `partial_output` attributes will be available.
     To make another attribute available, create a `TemporalRunContext` subclass with a custom `serialize_run_context` class method that returns a dictionary that includes the attribute and pass it to [`TemporalAgent`][pydantic_ai.durable_exec.temporal.TemporalAgent].
     """
 
@@ -44,9 +49,10 @@ class TemporalRunContext(RunContext[AgentDepsT]):
             'retry': ctx.retry,
             'max_retries': ctx.max_retries,
             'run_step': ctx.run_step,
+            'partial_output': ctx.partial_output,
         }
 
     @classmethod
-    def deserialize_run_context(cls, ctx: dict[str, Any], deps: AgentDepsT) -> TemporalRunContext[AgentDepsT]:
+    def deserialize_run_context(cls, ctx: dict[str, Any], deps: Any) -> TemporalRunContext[Any]:
         """Deserialize the run context from a `dict[str, Any]`."""
         return cls(**ctx, deps=deps)

pydantic_ai/durable_exec/temporal/_toolset.py

@@ -1,17 +1,58 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
-from collections.abc import Callable
-from typing import Any, Literal
+from collections.abc import Awaitable, Callable
+from dataclasses import dataclass
+from typing import Annotated, Any, Literal
 
+from pydantic import ConfigDict, Discriminator, with_config
 from temporalio.workflow import ActivityConfig
+from typing_extensions import assert_never
 
 from pydantic_ai import AbstractToolset, FunctionToolset, WrapperToolset
-from pydantic_ai.tools import AgentDepsT
+from pydantic_ai.exceptions import ApprovalRequired, CallDeferred, ModelRetry
+from pydantic_ai.tools import AgentDepsT, ToolDefinition
 
 from ._run_context import TemporalRunContext
 
 
+@dataclass
+@with_config(ConfigDict(arbitrary_types_allowed=True))
+class CallToolParams:
+    name: str
+    tool_args: dict[str, Any]
+    serialized_run_context: Any
+    tool_def: ToolDefinition | None
+
+
+@dataclass
+class _ApprovalRequired:
+    kind: Literal['approval_required'] = 'approval_required'
+
+
+@dataclass
+class _CallDeferred:
+    kind: Literal['call_deferred'] = 'call_deferred'
+
+
+@dataclass
+class _ModelRetry:
+    message: str
+    kind: Literal['model_retry'] = 'model_retry'
+
+
+@dataclass
+class _ToolReturn:
+    result: Any
+    kind: Literal['tool_return'] = 'tool_return'
+
+
+CallToolResult = Annotated[
+    _ApprovalRequired | _CallDeferred | _ModelRetry | _ToolReturn,
+    Discriminator('kind'),
+]
+
+
 class TemporalWrapperToolset(WrapperToolset[AgentDepsT], ABC):
     @property
     def id(self) -> str:
@@ -30,6 +71,29 @@ class TemporalWrapperToolset(WrapperToolset[AgentDepsT], ABC):
         # Temporalized toolsets cannot be swapped out after the fact.
         return self
 
+    async def _wrap_call_tool_result(self, coro: Awaitable[Any]) -> CallToolResult:
+        try:
+            result = await coro
+            return _ToolReturn(result=result)
+        except ApprovalRequired:
+            return _ApprovalRequired()
+        except CallDeferred:
+            return _CallDeferred()
+        except ModelRetry as e:
+            return _ModelRetry(message=e.message)
+
+    def _unwrap_call_tool_result(self, result: CallToolResult) -> Any:
+        if isinstance(result, _ToolReturn):
+            return result.result
+        elif isinstance(result, _ApprovalRequired):
+            raise ApprovalRequired()
+        elif isinstance(result, _CallDeferred):
+            raise CallDeferred()
+        elif isinstance(result, _ModelRetry):
+            raise ModelRetry(result.message)
+        else:
+            assert_never(result)
+
 
 def temporalize_toolset(
     toolset: AbstractToolset[AgentDepsT],

pydantic_ai/messages.py

@@ -13,7 +13,7 @@ import pydantic
 import pydantic_core
 from genai_prices import calc_price, types as genai_types
 from opentelemetry._events import Event  # pyright: ignore[reportPrivateImportUsage]
-from typing_extensions import Self, deprecated
+from typing_extensions import deprecated
 
 from . import _otel_messages, _utils
 from ._utils import generate_tool_call_id as _generate_tool_call_id, now_utc as _now_utc
@@ -34,6 +34,7 @@ DocumentMediaType: TypeAlias = Literal[
     'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet',
     'text/html',
     'text/markdown',
+    'application/msword',
     'application/vnd.ms-excel',
 ]
 VideoMediaType: TypeAlias = Literal[
@@ -434,8 +435,12 @@ class DocumentUrl(FileUrl):
             return 'application/pdf'
         elif self.url.endswith('.rtf'):
             return 'application/rtf'
+        elif self.url.endswith('.doc'):
+            return 'application/msword'
         elif self.url.endswith('.docx'):
             return 'application/vnd.openxmlformats-officedocument.wordprocessingml.document'
+        elif self.url.endswith('.xls'):
+            return 'application/vnd.ms-excel'
         elif self.url.endswith('.xlsx'):
             return 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet'
 
@@ -514,16 +519,16 @@ class BinaryContent:
                 vendor_metadata=bc.vendor_metadata,
             )
         else:
-            return bc  # pragma: no cover
+            return bc
 
     @classmethod
-    def from_data_uri(cls, data_uri: str) -> Self:
+    def from_data_uri(cls, data_uri: str) -> BinaryContent:
         """Create a `BinaryContent` from a data URI."""
         prefix = 'data:'
         if not data_uri.startswith(prefix):
-            raise ValueError('Data URI must start with "data:"')  # pragma: no cover
+            raise ValueError('Data URI must start with "data:"')
         media_type, data = data_uri[len(prefix) :].split(';base64,', 1)
-        return cls(data=base64.b64decode(data), media_type=media_type)
+        return cls.narrow_type(cls(data=base64.b64decode(data), media_type=media_type))
 
     @pydantic.computed_field
     @property
@@ -645,6 +650,7 @@ _document_format_lookup: dict[str, DocumentFormat] = {
     'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet': 'xlsx',
     'text/html': 'html',
     'text/markdown': 'md',
+    'application/msword': 'doc',
     'application/vnd.ms-excel': 'xls',
 }
 _audio_format_lookup: dict[str, AudioFormat] = {
@@ -882,7 +888,10 @@ class RetryPromptPart:
                 description = self.content
         else:
             json_errors = error_details_ta.dump_json(self.content, exclude={'__all__': {'ctx'}}, indent=2)
-            description = f'{len(self.content)} validation errors: {json_errors.decode()}'
+            plural = isinstance(self.content, list) and len(self.content) != 1
+            description = (
+                f'{len(self.content)} validation error{"s" if plural else ""}:\n```json\n{json_errors.decode()}\n```'
+            )
         return f'{description}\n\nFix the errors and try again.'
 
     def otel_event(self, settings: InstrumentationSettings) -> Event:
@@ -1612,6 +1621,14 @@ class PartStartEvent:
     part: ModelResponsePart
     """The newly started `ModelResponsePart`."""
 
+    previous_part_kind: (
+        Literal['text', 'thinking', 'tool-call', 'builtin-tool-call', 'builtin-tool-return', 'file'] | None
+    ) = None
+    """The kind of the previous part, if any.
+
+    This is useful for UI event streams to know whether to group parts of the same kind together when emitting events.
+    """
+
     event_kind: Literal['part_start'] = 'part_start'
     """Event type identifier, used as a discriminator."""
 
@@ -1634,6 +1651,30 @@ class PartDeltaEvent:
     __repr__ = _utils.dataclasses_no_defaults_repr
 
 
+@dataclass(repr=False, kw_only=True)
+class PartEndEvent:
+    """An event indicating that a part is complete."""
+
+    index: int
+    """The index of the part within the overall response parts list."""
+
+    part: ModelResponsePart
+    """The complete `ModelResponsePart`."""
+
+    next_part_kind: (
+        Literal['text', 'thinking', 'tool-call', 'builtin-tool-call', 'builtin-tool-return', 'file'] | None
+    ) = None
+    """The kind of the next part, if any.
+
+    This is useful for UI event streams to know whether to group parts of the same kind together when emitting events.
+    """
+
+    event_kind: Literal['part_end'] = 'part_end'
+    """Event type identifier, used as a discriminator."""
+
+    __repr__ = _utils.dataclasses_no_defaults_repr
+
+
 @dataclass(repr=False, kw_only=True)
 class FinalResultEvent:
     """An event indicating the response to the current model request matches the output schema and will produce a result."""
@@ -1649,9 +1690,9 @@ class FinalResultEvent:
 
 
 ModelResponseStreamEvent = Annotated[
-    PartStartEvent | PartDeltaEvent | FinalResultEvent, pydantic.Discriminator('event_kind')
+    PartStartEvent | PartDeltaEvent | PartEndEvent | FinalResultEvent, pydantic.Discriminator('event_kind')
 ]
-"""An event in the model response stream, starting a new part, applying a delta to an existing one, or indicating the final result."""
+"""An event in the model response stream, starting a new part, applying a delta to an existing one, indicating a part is complete, or indicating the final result."""
 
 
 @dataclass(repr=False)

pydantic_ai/models/__init__.py

@@ -27,6 +27,7 @@ from .._run_context import RunContext
 from ..builtin_tools import AbstractBuiltinTool
 from ..exceptions import UserError
 from ..messages import (
+    BaseToolCallPart,
     BinaryImage,
     FilePart,
     FileUrl,
@@ -35,9 +36,12 @@ from ..messages import (
     ModelMessage,
     ModelRequest,
     ModelResponse,
+    ModelResponsePart,
     ModelResponseStreamEvent,
+    PartEndEvent,
     PartStartEvent,
     TextPart,
+    ThinkingPart,
     ToolCallPart,
     VideoUrl,
 )
@@ -543,7 +547,44 @@ class StreamedResponse(ABC):
                 async for event in iterator:
                     yield event
 
-            self._event_iterator = iterator_with_final_event(self._get_event_iterator())
+            async def iterator_with_part_end(
+                iterator: AsyncIterator[ModelResponseStreamEvent],
+            ) -> AsyncIterator[ModelResponseStreamEvent]:
+                last_start_event: PartStartEvent | None = None
+
+                def part_end_event(next_part: ModelResponsePart | None = None) -> PartEndEvent | None:
+                    if not last_start_event:
+                        return None
+
+                    index = last_start_event.index
+                    part = self._parts_manager.get_parts()[index]
+                    if not isinstance(part, TextPart | ThinkingPart | BaseToolCallPart):
+                        # Parts other than these 3 don't have deltas, so don't need an end part.
+                        return None
+
+                    return PartEndEvent(
+                        index=index,
+                        part=part,
+                        next_part_kind=next_part.part_kind if next_part else None,
+                    )
+
+                async for event in iterator:
+                    if isinstance(event, PartStartEvent):
+                        if last_start_event:
+                            end_event = part_end_event(event.part)
+                            if end_event:
+                                yield end_event
+
+                            event.previous_part_kind = last_start_event.part.part_kind
+                        last_start_event = event
+
+                    yield event
+
+                end_event = part_end_event()
+                if end_event:
+                    yield end_event
+
+            self._event_iterator = iterator_with_part_end(iterator_with_final_event(self._get_event_iterator()))
         return self._event_iterator
 
     @abstractmethod

pydantic_ai/models/google.py

@@ -471,11 +471,9 @@ class GoogleModel(Model):
                 raise UnexpectedModelBehavior(
                     f'Content filter {raw_finish_reason.value!r} triggered', response.model_dump_json()
                 )
-            else:
-                raise UnexpectedModelBehavior(
-                    'Content field missing from Gemini response', response.model_dump_json()
-                )  # pragma: no cover
-        parts = candidate.content.parts or []
+            parts = []  # pragma: no cover
+        else:
+            parts = candidate.content.parts or []
 
         usage = _metadata_as_usage(response)
         return _process_response_from_parts(
@@ -649,17 +647,12 @@ class GeminiStreamedResponse(StreamedResponse):
             #     )
 
             if candidate.content is None or candidate.content.parts is None:
-                if self.finish_reason == 'stop':  # pragma: no cover
-                    # Normal completion - skip this chunk
-                    continue
-                elif self.finish_reason == 'content_filter' and raw_finish_reason:  # pragma: no cover
+                if self.finish_reason == 'content_filter' and raw_finish_reason:  # pragma: no cover
                     raise UnexpectedModelBehavior(
                         f'Content filter {raw_finish_reason.value!r} triggered', chunk.model_dump_json()
                     )
                 else:  # pragma: no cover
-                    raise UnexpectedModelBehavior(
-                        'Content field missing from streaming Gemini response', chunk.model_dump_json()
-                    )
+                    continue
 
             parts = candidate.content.parts
             if not parts:

pydantic_ai/models/groq.py

@@ -524,6 +524,8 @@ class GroqStreamedResponse(StreamedResponse):
     async def _get_event_iterator(self) -> AsyncIterator[ModelResponseStreamEvent]:  # noqa: C901
         try:
             executed_tool_call_id: str | None = None
+            reasoning_index = 0
+            reasoning = False
             async for chunk in self._response:
                 self._usage += _map_usage(chunk)
 
@@ -540,10 +542,16 @@ class GroqStreamedResponse(StreamedResponse):
                     self.finish_reason = _FINISH_REASON_MAP.get(raw_finish_reason)
 
                 if choice.delta.reasoning is not None:
+                    if not reasoning:
+                        reasoning_index += 1
+                        reasoning = True
+
                     # NOTE: The `reasoning` field is only present if `groq_reasoning_format` is set to `parsed`.
                     yield self._parts_manager.handle_thinking_delta(
-                        vendor_part_id='reasoning', content=choice.delta.reasoning
+                        vendor_part_id=f'reasoning-{reasoning_index}', content=choice.delta.reasoning
                     )
+                else:
+                    reasoning = False
 
                 if choice.delta.executed_tools:
                     for tool in choice.delta.executed_tools:

pydantic_ai/models/openai.py

@@ -948,6 +948,10 @@ class OpenAIResponsesModel(Model):
 
         super().__init__(settings=settings, profile=profile or provider.model_profile)
 
+    @property
+    def base_url(self) -> str:
+        return str(self.client.base_url)
+
     @property
     def model_name(self) -> OpenAIModelName:
         """The model name."""
@@ -1148,10 +1152,10 @@ class OpenAIResponsesModel(Model):
             + list(model_settings.get('openai_builtin_tools', []))
             + self._get_tools(model_request_parameters)
         )
-
+        profile = OpenAIModelProfile.from_profile(self.profile)
         if not tools:
             tool_choice: Literal['none', 'required', 'auto'] | None = None
-        elif not model_request_parameters.allow_text_output:
+        elif not model_request_parameters.allow_text_output and profile.openai_supports_tool_choice_required:
             tool_choice = 'required'
         else:
             tool_choice = 'auto'
@@ -1184,7 +1188,6 @@ class OpenAIResponsesModel(Model):
             text = text or {}
             text['verbosity'] = verbosity
 
-        profile = OpenAIModelProfile.from_profile(self.profile)
         unsupported_model_settings = profile.openai_unsupported_model_settings
         for setting in unsupported_model_settings:
             model_settings.pop(setting, None)

pydantic_ai/profiles/openai.py

@@ -62,7 +62,10 @@ class OpenAIModelProfile(ModelProfile):
 
 def openai_model_profile(model_name: str) -> ModelProfile:
     """Get the model profile for an OpenAI model."""
-    is_reasoning_model = model_name.startswith('o') or model_name.startswith('gpt-5')
+    is_gpt_5 = model_name.startswith('gpt-5')
+    is_o_series = model_name.startswith('o')
+    is_reasoning_model = is_o_series or (is_gpt_5 and 'gpt-5-chat' not in model_name)
+
     # Check if the model supports web search (only specific search-preview models)
     supports_web_search = '-search-preview' in model_name
 
@@ -91,7 +94,7 @@ def openai_model_profile(model_name: str) -> ModelProfile:
         json_schema_transformer=OpenAIJsonSchemaTransformer,
         supports_json_schema_output=True,
         supports_json_object_output=True,
-        supports_image_output=is_reasoning_model or '4.1' in model_name or '4o' in model_name,
+        supports_image_output=is_gpt_5 or 'o3' in model_name or '4.1' in model_name or '4o' in model_name,
         openai_unsupported_model_settings=openai_unsupported_model_settings,
         openai_system_prompt_role=openai_system_prompt_role,
         openai_chat_supports_web_search=supports_web_search,