Skip to content

feat: add gemini-3-flash-preview model#96

Merged
Kamilbenkirane merged 1 commit intomainfrom
feat/add-gemini-3-flash-preview
Dec 22, 2025
Merged

feat: add gemini-3-flash-preview model#96
Kamilbenkirane merged 1 commit intomainfrom
feat/add-gemini-3-flash-preview

Conversation

@Kamilbenkirane
Copy link
Copy Markdown
Member

@Kamilbenkirane Kamilbenkirane commented Dec 22, 2025

Adds Gemini 3 Flash Preview model to Google text generation models.

This model supports:

  • Thinking level parameter (low/high)
  • Output schema

Related to: https://withceleste.ai/models/gemini-3-flash-preview

Add Gemini 3 Flash Preview model to Google text generation models.
@claude
Copy link
Copy Markdown

claude Bot commented Dec 22, 2025

Pull Request Review: Add Gemini 3 Flash Preview Model

Thank you for the contribution! I've reviewed the changes and have the following feedback:

✅ Code Quality & Best Practices

Positive:

  • The code follows the established pattern used for other Google models in the file
  • Consistent formatting and structure with existing model definitions
  • Proper use of type hints and constraints
  • Clean, readable implementation

Observations:

  • The model configuration closely mirrors gemini-3-pro-preview (lines 53-63), which is good for consistency
  • All required fields (id, provider, display_name, streaming, parameter_constraints) are properly defined

🔍 Parameter Configuration Review

Temperature: Range(min=0.0, max=2.0)

  • Consistent with other Gemini models
  • Standard range for Google models

Max Tokens: Range(min=1, max=65536)

  • Matches other Gemini 3/2.5 models
  • Appropriate maximum token limit

Thinking Level: Choice(options=["low", "high"])

  • Identical to gemini-3-pro-preview, which is correct for Gemini 3 series
  • Note: This differs from Gemini 2.5 models which use THINKING_BUDGET with numeric ranges
  • Matches the claimed capabilities in the PR description

Output Schema: Schema()

  • Properly included, consistent with other capable models

⚠️ Concerns & Recommendations

1. Missing Tests
The integration tests at packages/capabilities/text-generation/tests/integration_tests/test_text_generation/test_generate.py do not include the new model. Consider adding a test case:

(Provider.GOOGLE, "gemini-3-flash-preview", {"thinking_level": "low"}),

2. Capabilities Verification
The PR description mentions several capabilities:

  • ✅ Text generation - supported
  • ✅ Thinking level parameter - supported
  • ✅ Output schema - supported
  • ❓ Audio intelligence - not explicitly configured
  • ❓ Image intelligence - not explicitly configured
  • ❓ Video intelligence - not explicitly configured
  • ❓ Web search - not explicitly configured

Question: Are the multimodal capabilities (audio, image, video) and web search handled at a different layer (e.g., in the Google provider implementation), or should they be reflected in the parameter constraints? If they're handled elsewhere, the current implementation is fine. If not, we may need additional parameters.

3. Documentation
Consider updating integration tests or documentation to include this new model in examples, especially if it has unique capabilities compared to other models.

🔒 Security Concerns

No security concerns identified. The change is minimal and only adds metadata configuration.

⚡ Performance Considerations

No performance impact expected. This is a configuration-only change.

📊 Overall Assessment

Verdict:Approve with minor suggestions

The code quality is good and follows established patterns. The main recommendation is to add integration test coverage. The multimodal capabilities mentioned in the PR description should be clarified - either confirm they're handled elsewhere or update the configuration accordingly.

Suggested Next Steps

  1. Add integration test case for the new model
  2. Clarify where multimodal and web search capabilities are configured
  3. Consider adding the model to the README or documentation if it has unique features worth highlighting

Great work overall! 🚀

@Kamilbenkirane Kamilbenkirane merged commit d21bb2f into main Dec 22, 2025
11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant