Gemini models are not seeded

[khalidMindee]

Hi,

Gemini models support seed in the generationConfig parameter , but when parsing the config it's not taken , the seed parameter is not set from the input model_settings, is that intentional ?

pydantic-ai/pydantic_ai_slim/pydantic_ai/models/google.py

Lines 391 to 409 in 0047a68

	config = GenerateContentConfigDict(
	http_options=http_options,
	system_instruction=system_instruction,
	temperature=model_settings.get('temperature'),
	top_p=model_settings.get('top_p'),
	max_output_tokens=model_settings.get('max_tokens'),
	stop_sequences=model_settings.get('stop_sequences'),
	presence_penalty=model_settings.get('presence_penalty'),
	frequency_penalty=model_settings.get('frequency_penalty'),
	safety_settings=model_settings.get('google_safety_settings'),
	thinking_config=model_settings.get('google_thinking_config'),
	labels=model_settings.get('google_labels'),
	media_resolution=model_settings.get('google_video_resolution'),
	cached_content=model_settings.get('google_cached_content'),
	tools=cast(ToolListUnionDict, tools),
	tool_config=tool_config,
	response_mime_type=response_mime_type,
	response_schema=response_schema,
	)

Thanks,

[DouweM]

@khalidMindee Not intentional, let's add it.

@claude Please submit a PR that makes GoogleModel respect the seed field from ModelSettings. Don't forget to update the ModelSettings docstring, and also pass the seed in GoogleModel.count_tokens.

[claude]

Claude finished @DouweM's task —— View job • claude/issue-2838-20250909-1526 • Create PR ➔

---

✅ Successfully added seed support to GoogleModel

I've implemented seed parameter support for Gemini models as requested. Here's what was completed:

• Set up development environment
• Analyze current GoogleModel implementation
• Check how other models handle seed parameter
• Implement seed support in GoogleModel
• Update ModelSettings docstring
• Update count_tokens method to pass seed
• Run tests and linting
• Commit and push changes
• Provide PR creation URL

Changes Made:

1. Added seed parameter to GoogleModel: Modified _build_content_and_config method to include seed=model_settings.get('seed') in the GenerateContentConfigDict

2. Updated count_tokens method: Added seed parameter to the GenerationConfigDict used in the count_tokens method for Vertex AI support

3. Updated documentation: Added "Gemini" to the list of supported providers for the seed parameter in ModelSettings docstring

Testing:

• ✅ All linting and formatting checks pass
• ✅ Google model tests pass (57/59 - the 2 failures are unrelated network issues with Anthropic API)
• ✅ No regressions introduced