PR #27007: fix(core): add aliases and thinking config for gemini-3.1 models

• Repository: google-gemini/gemini-cli
• Author: anishs1207
• State: open | merged: False
• Link: https://github.com/google-gemini/gemini-cli/pull/27007

Description (problem / solution / changelog)

Summary

This PR resolves the INVALID_ARGUMENT (400) API error encountered when using Gemini 3.1 models. It ensures that all Gemini 3.1 variants are correctly mapped to the chat-base-3 configuration, which includes the mandatory thinkingLevel parameter required by the Google Generative AI API.

Details

The investigation revealed that while gemini-3.1-pro-preview was present, it lacked an entry in the aliases section of defaultModelConfigs.ts. This caused the CLI to fall back to a basic configuration that enabled "Thinking" mode but failed to provide a thinkingLevel (e.g., HIGH). Gemini 3.1 models strictly require this level when thinking is enabled; without it, the API rejects the request as an invalid argument.

Changes include:

• Added gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools, and gemini-3.1-flash-lite-preview to the core alias registry.
• Mapped these models to inherit from chat-base-3 to ensure ThinkingLevel.HIGH is included in all requests.

Related Issues

Fixes #26997, #27000, #27005, #26996

How to Validate

1. Run Tests: Execute the updated test suite in packages/core/src/config/models.test.ts to ensure 3.1 models resolve with the correct config.
   • npm run test core
2. Manual Verification: Use a Gemini 3.1 model to process a comparison prompt (e.g., "AMD Ryzen 9 9950X vs 9950X3D") and verify the 400 error no longer occurs.

Pre-Merge Checklist

• Updated relevant documentation and README (if needed)
• Added/updated tests (if needed)
• Noted breaking changes (if any)
• Validated on required platforms/methods:
  • Windows (Validated via MINGW64/git-bash)
    • npm run

Changed files

• packages/core/src/config/defaultModelConfigs.ts (modified, +18/-0)
• packages/core/src/config/models.test.ts (modified, +32/-0)