Skip to content

Warmup request downgraded to model: gpt-5-mini - Claude Code

Quick fix

Set the Copilot optimizer's warmupDowngrade to false to prevent warmup requests from being rewritten to the unavailable gpt-5-mini.

Error output
[Copilot] Warmup request downgraded to model: gpt-5-mini

The CC Switch Copilot optimizer enables warmup downgrade by default and uses `gpt-5-mini` as the fallback model. After a request is classified as warmup, the forwarder overrides the already resolved Claude model. As a result, the explicitly selected `claude-opus-4.8-1m` is first resolved to `claude-opus-4.8`, then silently rewritten to `gpt-5-mini`.

The issue was reproduced on macOS with CC Switch 3.18.0 and 4.0.10. The 4.0.10 report also indicates that Claude Desktop security classification and model availability checks may be misclassified as warmup. If the Copilot subscription does not support `gpt-5-mini`, the upstream service returns HTTP 400 and `model_not_supported`. The issue was eventually closed due to inactivity. The maintainer recommended upgrading to the latest version and trying again, but the thread did not identify a fixed version.

  1. Set `warmupDowngrade` to `false` in the Copilot optimizer configuration.

  2. Restart or use a Claude Code session and inspect the logs. Confirm that after live-model resolves `claude-opus-4.8-1m` to `claude-opus-4.8`, it is no longer replaced with `gpt-5-mini`.

  3. If the issue remains reproducible on the latest version, provide the CC Switch version, operating system, complete reproduction steps, and relevant logs or screenshots, then reopen the issue.

ToolClaude Code
Version受影响版本:3.18.0 - 4.0.10
PlatformsmacOS
Does this issue affect only Claude Code?
No. A reproduction report in the thread says Claude Desktop used through the proxy gateway is also affected on CC Switch 4.0.10, including security reviews before built-in browser actions and availability checks when switching models.
Why does GitHub Copilot return `model_not_supported`?
The reported GitHub Copilot Business subscription provides only Claude models and does not include `gpt-5-mini`. After the warmup optimizer rewrites the request to `gpt-5-mini`, the upstream service returns HTTP 400 and `model_not_supported`.
Was this issue fixed after the 4.0 refactor?
This cannot be confirmed. The maintainer said 4.0 substantially refactored the relevant functionality and recommended updating to the latest version, but the user still reproduced the issue on 4.0.10. The thread does not identify a fixed version.

This is an unofficial community wiki with no affiliation to the cc-switch authors or the project itself. Its content is compiled from the project's public GitHub issues. This site distributes no software.