Merge branch 'main' into fix/issue-2992-disallow-initial-model-response

Author: DouweM

docs/models/google.md

@@ -214,22 +214,22 @@ from pydantic_ai.models.google import GoogleModel, GoogleModelSettings
 settings = GoogleModelSettings(
     temperature=0.2,
     max_tokens=1024,
-    google_thinking_config={'thinking_budget': 2048},
+    google_thinking_config={'thinking_level': 'low'},
     google_safety_settings=[
         {
             'category': HarmCategory.HARM_CATEGORY_HATE_SPEECH,
             'threshold': HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
         }
     ]
 )
-model = GoogleModel('gemini-2.5-flash')
+model = GoogleModel('gemini-2.5-pro')
 agent = Agent(model, model_settings=settings)
 ...
 ```
 
 ### Disable thinking
 
-You can disable thinking by setting the `thinking_budget` to `0` on the `google_thinking_config`:
+On models older than Gemini 2.5 Pro, you can disable thinking by setting the `thinking_budget` to `0` on the `google_thinking_config`:
 
 ```python
 from pydantic_ai import Agent

pydantic_ai_slim/pydantic_ai/models/__init__.py

@@ -145,24 +145,28 @@
         'cohere:command-r7b-12-2024',
         'deepseek:deepseek-chat',
         'deepseek:deepseek-reasoner',
+        'google-gla:gemini-flash-latest',
+        'google-gla:gemini-flash-lite-latest',
         'google-gla:gemini-2.0-flash',
         'google-gla:gemini-2.0-flash-lite',
         'google-gla:gemini-2.5-flash',
+        'google-gla:gemini-2.5-flash-preview-09-2025',
+        'google-gla:gemini-2.5-flash-image',
         'google-gla:gemini-2.5-flash-lite',
         'google-gla:gemini-2.5-flash-lite-preview-09-2025',
-        'google-gla:gemini-2.5-flash-preview-09-2025',
         'google-gla:gemini-2.5-pro',
-        'google-gla:gemini-flash-latest',
-        'google-gla:gemini-flash-lite-latest',
+        'google-gla:gemini-3-pro-preview',
+        'google-vertex:gemini-flash-latest',
+        'google-vertex:gemini-flash-lite-latest',
         'google-vertex:gemini-2.0-flash',
         'google-vertex:gemini-2.0-flash-lite',
         'google-vertex:gemini-2.5-flash',
+        'google-vertex:gemini-2.5-flash-preview-09-2025',
+        'google-vertex:gemini-2.5-flash-image',
         'google-vertex:gemini-2.5-flash-lite',
         'google-vertex:gemini-2.5-flash-lite-preview-09-2025',
-        'google-vertex:gemini-2.5-flash-preview-09-2025',
         'google-vertex:gemini-2.5-pro',
-        'google-vertex:gemini-flash-latest',
-        'google-vertex:gemini-flash-lite-latest',
+        'google-vertex:gemini-3-pro-preview',
         'grok:grok-2-image-1212',
         'grok:grok-2-vision-1212',
         'grok:grok-3',

pydantic_ai_slim/pydantic_ai/models/google.py

@@ -14,7 +14,7 @@
 from .._output import OutputObjectDefinition
 from .._run_context import RunContext
 from ..builtin_tools import CodeExecutionTool, ImageGenerationTool, UrlContextTool, WebSearchTool
-from ..exceptions import UserError
+from ..exceptions import ModelHTTPError, UserError
 from ..messages import (
     BinaryContent,
     BuiltinToolCallPart,
@@ -38,6 +38,7 @@
     VideoUrl,
 )
 from ..profiles import ModelProfileSpec
+from ..profiles.google import GoogleModelProfile
 from ..providers import Provider, infer_provider
 from ..settings import ModelSettings
 from ..tools import ToolDefinition
@@ -51,7 +52,7 @@
 )
 
 try:
-    from google.genai import Client
+    from google.genai import Client, errors
     from google.genai.types import (
         BlobDict,
         CodeExecutionResult,
@@ -93,15 +94,17 @@
     ) from _import_error
 
 LatestGoogleModelNames = Literal[
+    'gemini-flash-latest',
+    'gemini-flash-lite-latest',
     'gemini-2.0-flash',
     'gemini-2.0-flash-lite',
     'gemini-2.5-flash',
     'gemini-2.5-flash-preview-09-2025',
-    'gemini-flash-latest',
+    'gemini-2.5-flash-image',
     'gemini-2.5-flash-lite',
     'gemini-2.5-flash-lite-preview-09-2025',
-    'gemini-flash-lite-latest',
     'gemini-2.5-pro',
+    'gemini-3-pro-preview',
 ]
 """Latest Gemini models."""
 
@@ -228,12 +231,17 @@ def system(self) -> str:
     def prepare_request(
         self, model_settings: ModelSettings | None, model_request_parameters: ModelRequestParameters
     ) -> tuple[ModelSettings | None, ModelRequestParameters]:
+        supports_native_output_with_builtin_tools = GoogleModelProfile.from_profile(
+            self.profile
+        ).google_supports_native_output_with_builtin_tools
         if model_request_parameters.builtin_tools and model_request_parameters.output_tools:
             if model_request_parameters.output_mode == 'auto':
-                model_request_parameters = replace(model_request_parameters, output_mode='prompted')
+                output_mode = 'native' if supports_native_output_with_builtin_tools else 'prompted'
+                model_request_parameters = replace(model_request_parameters, output_mode=output_mode)
             else:
+                output_mode = 'NativeOutput' if supports_native_output_with_builtin_tools else 'PromptedOutput'
                 raise UserError(
-                    'Google does not support output tools and built-in tools at the same time. Use `output_type=PromptedOutput(...)` instead.'
+                    f'Google does not support output tools and built-in tools at the same time. Use `output_type={output_mode}(...)` instead.'
                 )
         return super().prepare_request(model_settings, model_request_parameters)
 
@@ -394,7 +402,16 @@ async def _generate_content(
     ) -> GenerateContentResponse | Awaitable[AsyncIterator[GenerateContentResponse]]:
         contents, config = await self._build_content_and_config(messages, model_settings, model_request_parameters)
         func = self.client.aio.models.generate_content_stream if stream else self.client.aio.models.generate_content
-        return await func(model=self._model_name, contents=contents, config=config)  # type: ignore
+        try:
+            return await func(model=self._model_name, contents=contents, config=config)  # type: ignore
+        except errors.APIError as e:
+            if (status_code := e.code) >= 400:
+                raise ModelHTTPError(
+                    status_code=status_code,
+                    model_name=self._model_name,
+                    body=cast(Any, e.details),  # pyright: ignore[reportUnknownMemberType]
+                ) from e
+            raise  # pragma: lax no cover
 
     async def _build_content_and_config(
         self,
@@ -409,9 +426,9 @@ async def _build_content_and_config(
         response_mime_type = None
         response_schema = None
         if model_request_parameters.output_mode == 'native':
-            if tools:
+            if model_request_parameters.function_tools:
                 raise UserError(
-                    'Google does not support `NativeOutput` and tools at the same time. Use `output_type=ToolOutput(...)` instead.'
+                    'Google does not support `NativeOutput` and function tools at the same time. Use `output_type=ToolOutput(...)` instead.'
                 )
             response_mime_type = 'application/json'
             output_object = model_request_parameters.output_object
@@ -676,23 +693,25 @@ async def _get_event_iterator(self) -> AsyncIterator[ModelResponseStreamEvent]:
 
             for part in parts:
                 if part.thought_signature:
+                    # Per https://ai.google.dev/gemini-api/docs/function-calling?example=meeting#thought-signatures:
+                    # - Always send the thought_signature back to the model inside its original Part.
+                    # - Don't merge a Part containing a signature with one that does not. This breaks the positional context of the thought.
+                    # - Don't combine two Parts that both contain signatures, as the signature strings cannot be merged.
+
                     signature = base64.b64encode(part.thought_signature).decode('utf-8')
+                    # Attach signature to most recent thinking part, if there was one
                     yield self._parts_manager.handle_thinking_delta(
-                        vendor_part_id='thinking',
+                        vendor_part_id=None,
                         signature=signature,
                         provider_name=self.provider_name,
                     )
 
                 if part.text is not None:
                     if len(part.text) > 0:
                         if part.thought:
-                            yield self._parts_manager.handle_thinking_delta(
-                                vendor_part_id='thinking', content=part.text
-                            )
+                            yield self._parts_manager.handle_thinking_delta(vendor_part_id=None, content=part.text)
                         else:
-                            maybe_event = self._parts_manager.handle_text_delta(
-                                vendor_part_id='content', content=part.text
-                            )
+                            maybe_event = self._parts_manager.handle_text_delta(vendor_part_id=None, content=part.text)
                             if maybe_event is not None:  # pragma: no branch
                                 yield maybe_event
                 elif part.function_call:
@@ -751,6 +770,7 @@ def timestamp(self) -> datetime:
 def _content_model_response(m: ModelResponse, provider_name: str) -> ContentDict:  # noqa: C901
     parts: list[PartDict] = []
     thought_signature: bytes | None = None
+    function_call_requires_signature: bool = True
     for item in m.parts:
         part: PartDict = {}
         if thought_signature:
@@ -760,6 +780,15 @@ def _content_model_response(m: ModelResponse, provider_name: str) -> ContentDict
         if isinstance(item, ToolCallPart):
             function_call = FunctionCallDict(name=item.tool_name, args=item.args_as_dict(), id=item.tool_call_id)
             part['function_call'] = function_call
+            if function_call_requires_signature and not part.get('thought_signature'):
+                # Per https://ai.google.dev/gemini-api/docs/gemini-3?thinking=high#migrating_from_other_models:
+                # > If you are transferring a conversation trace from another model (e.g., Gemini 2.5) or injecting
+                # > a custom function call that was not generated by Gemini 3, you will not have a valid signature.
+                # > To bypass strict validation in these specific scenarios, populate the field with this specific
+                # > dummy string: "thoughtSignature": "context_engineering_is_the_way_to_go"
+                part['thought_signature'] = b'context_engineering_is_the_way_to_go'
+            # Only the first function call requires a signature
+            function_call_requires_signature = False
         elif isinstance(item, TextPart):
             part['text'] = item.content
         elif isinstance(item, ThinkingPart):
@@ -872,7 +901,7 @@ def _function_declaration_from_tool(tool: ToolDefinition) -> FunctionDeclaration
     f = FunctionDeclarationDict(
         name=tool.name,
         description=tool.description or '',
-        parameters=json_schema,  # type: ignore
+        parameters_json_schema=json_schema,
     )
     return f
 

pydantic_ai_slim/pydantic_ai/profiles/google.py

@@ -1,18 +1,34 @@
 from __future__ import annotations as _annotations
 
+from dataclasses import dataclass
+
 from .._json_schema import JsonSchema, JsonSchemaTransformer
 from . import ModelProfile
 
 
+@dataclass(kw_only=True)
+class GoogleModelProfile(ModelProfile):
+    """Profile for models used with `GoogleModel`.
+
+    ALL FIELDS MUST BE `google_` PREFIXED SO YOU CAN MERGE THEM WITH OTHER MODELS.
+    """
+
+    google_supports_native_output_with_builtin_tools: bool = False
+    """Whether the model supports native output with builtin tools.
+    See https://ai.google.dev/gemini-api/docs/structured-output?example=recipe#structured_outputs_with_tools"""
+
+
 def google_model_profile(model_name: str) -> ModelProfile | None:
     """Get the model profile for a Google model."""
     is_image_model = 'image' in model_name
-    return ModelProfile(
+    is_3_or_newer = 'gemini-3' in model_name
+    return GoogleModelProfile(
         json_schema_transformer=GoogleJsonSchemaTransformer,
         supports_image_output=is_image_model,
         supports_json_schema_output=not is_image_model,
         supports_json_object_output=not is_image_model,
         supports_tools=not is_image_model,
+        google_supports_native_output_with_builtin_tools=is_3_or_newer,
     )
 
 

pydantic_ai_slim/pyproject.toml

@@ -70,7 +70,7 @@ logfire = ["logfire[httpx]>=3.14.1"]
 openai = ["openai>=1.107.2"]
 cohere = ["cohere>=5.18.0; platform_system != 'Emscripten'"]
 vertexai = ["google-auth>=2.36.0", "requests>=2.32.2"]
-google = ["google-genai>=1.50.1"]
+google = ["google-genai>=1.51.0"]
 anthropic = ["anthropic>=0.70.0"]
 groq = ["groq>=0.25.0"]
 mistral = ["mistralai>=1.9.10"]

tests/models/cassettes/test_google/test_google_builtin_tools_with_other_tools.yaml

@@ -8,7 +8,7 @@ interactions:
       connection:
       - keep-alive
       content-length:
-      - '526'
+      - '560'
       content-type:
       - application/json
       host:
@@ -19,10 +19,13 @@ interactions:
       - parts:
         - text: What is the largest city in Mexico?
         role: user
-      generationConfig: {}
+      generationConfig:
+        responseModalities:
+        - TEXT
       systemInstruction:
         parts:
-        - text: |-
+        - text: |2
+
             Always respond with a JSON object that's compatible with this schema:
 
             {"properties": {"city": {"type": "string"}, "country": {"type": "string"}}, "required": ["city", "country"], "title": "CityLocation", "type": "object"}
@@ -41,7 +44,7 @@ interactions:
       content-type:
       - application/json; charset=UTF-8
       server-timing:
-      - gfet4t7; dur=780
+      - gfet4t7; dur=873
       transfer-encoding:
       - chunked
       vary:
@@ -58,15 +61,15 @@ interactions:
         groundingMetadata: {}
         index: 0
       modelVersion: gemini-2.5-flash
-      responseId: 6Xq3aPnXNtqKqtsP8ZuDyAc
+      responseId: 0f4caf7iM-mgz7IPtLeryAc
       usageMetadata:
         candidatesTokenCount: 13
-        promptTokenCount: 83
+        promptTokenCount: 85
         promptTokensDetails:
         - modality: TEXT
-          tokenCount: 83
-        thoughtsTokenCount: 33
-        totalTokenCount: 129
+          tokenCount: 85
+        thoughtsTokenCount: 32
+        totalTokenCount: 130
     status:
       code: 200
       message: OK