mirror of
https://github.com/google-gemini/gemini-cli.git
synced 2026-06-25 18:56:48 -07:00
gemini-cli[bot]
f146618b6b
This PR implements auth-aware model routing to prevent 404/400 errors for users using personal OAuth (`oauth-personal`). ### Problem When using the `auto-gemini-3` model alias with personal OAuth, the CLI occasionally resolves it to `gemini-3-pro-preview`. However, many personal accounts do not have access to this specific preview model, leading to API errors (404/400). Currently, the CLI only falls back to stable models if the user has NO access to ANY preview models, which is too broad if they have access to older preview models but not the latest ones. ### Solution - Added `authType` to the `ModelResolutionContext` and `ModelCapabilityContext`. - Updated `ModelConfigService` to support conditions based on `authType`. - Modified `DEFAULT_MODEL_CONFIGS` and `settings.schema.json` to explicitly route `oauth-personal` users to stable models (e.g., `gemini-2.5-pro`) for `auto` and `pro` aliases. - Updated `resolveModel` to pass the current `authType` to the resolution engine. These changes ensure that personal OAuth users always get a working stable model by default, while API key and Vertex AI users can still access the latest previews. Fixes: #26857 cc @NTaylorMullen