What happened?
Environment
- @google/gemini-cli version: 0.56.0
- Node.js: v22.17.0
- OS: Windows 11 (win32)
- Configured model: gemini-3.1-pro-preview
- API endpoint: custom proxy (OpenAI/new-api compatible relay with model-routing "groups", group name cocos)
Description
I configured gemini-3.1-pro-preview as my model. Normal chat requests use this model correctly. However, when gemini-cli tries to generate a background session summary, it silently switches to gemini-3.1-flash-lite instead of the configured model. My proxy's cocos group has no channel/distributor for gemini-3.1-flash-lite, so every summary attempt fails with a model_not_found 503 error, retries 4+ times with backoff, and spams the terminal with stack traces.
Steps to reproduce
- Set model to gemini-3.1-pro-preview via /model (or config), pointed at a custom API base URL that only exposes specific models per routing group.
- Use the CLI normally until a session summary is triggered (e.g., on /model switch, compaction, or session end — whichever triggers generateAndSaveSummary).
- Observe the summary generation request goes to gemini-3.1-flash-lite, not the configured model.
What did you expect to happen?
Session summary generation should use the currently configured/active model (or at minimum, a model the user has explicitly made available), not a hardcoded lite model that may not exist on the user's endpoint.
Client information
Client Information
Run gemini to enter the interactive CLI, then run the /about command.
> /about
│ About Gemini CLI │
│ │
│ CLI Version 0.56.0 │
│ Git Commit b6e23a7dc │
│ Model gemini-3.1-pro-preview │
│ Sandbox no sandbox │
│ OS win32 │
│ Auth Method gemini-api-key
Login information
Attempt 5 failed: {"error":{"code":"model_not_found","message":"分组 cocos 下模型 gemini-3.1-flash-lite 无可用渠道（distributor） (request id: 202608250246317784272508xxxxxxx)","type":"new_api_error"}}. Max attempts reached
Anything else we need to know?
No response
What happened?
Environment
Description
I configured gemini-3.1-pro-preview as my model. Normal chat requests use this model correctly. However, when gemini-cli tries to generate a background session summary, it silently switches to gemini-3.1-flash-lite instead of the configured model. My proxy's cocos group has no channel/distributor for gemini-3.1-flash-lite, so every summary attempt fails with a model_not_found 503 error, retries 4+ times with backoff, and spams the terminal with stack traces.
Steps to reproduce
What did you expect to happen?
Session summary generation should use the currently configured/active model (or at minimum, a model the user has explicitly made available), not a hardcoded lite model that may not exist on the user's endpoint.
Client information
Client Information
Run
geminito enter the interactive CLI, then run the/aboutcommand.Login information
Attempt 5 failed: {"error":{"code":"model_not_found","message":"分组 cocos 下模型 gemini-3.1-flash-lite 无可用渠道（distributor） (request id: 202608250246317784272508xxxxxxx)","type":"new_api_error"}}. Max attempts reached
Anything else we need to know?
No response