Skip to content

Configuring DeepSeek in Codex errors about gpt-5.5 - CC Switch

Quick fix

Install the Codex CLI and run codex debug models --bundled to regenerate models_cache.json.

Error output
配置codex的时候,切换供应商或者是开启路由的时候报错,我配置的是deepseek,里面压根没有 gpt-5.5
English translation

Configuring codex, switching provider or turning routing on errors. I configured deepseek, which has no gpt-5.5 in it at all.

Switching the Codex provider, cc-switch depends on the Codex model cache file models_cache.json. If only the Codex desktop application is installed on your machine and the Codex CLI is not, ~/.codex/models_cache.json may not exist or may hold the wrong content, so switching to the DeepSeek provider still errors about a non-existent model such as gpt-5.5. The problem is related to Codex config sync and model catalog refresh. A neighbouring problem is the Codex base_url being rewritten after a cc-switch restart, which maintainers say was handled in v3.16.1 and newer along with fixes for Codex config/takeover switching and model catalog refresh.

  1. Confirm the Codex CLI is installed on your machine — installing only the Codex app does not generate models_cache.json.

  2. Run the command for your platform to regenerate the Codex model cache file.

    macOS / Linux
    codex debug models --bundled > ~/.codex/models_cache.json
    # Windows PowerShell
    codex debug models --bundled > $HOME\.codex\models_cache.json
    # Windows CMD
    codex debug models --bundled > %USERPROFILE%\.codex\models_cache.json
  3. Switch the Codex provider or turn routing on again in cc-switch, and confirm the DeepSeek configuration works.

  4. If you also hit the Codex base_url changing after a cc-switch restart and breaking things, upgrade cc-switch to v3.16.1 or newer and verify again.

ToolCodex
Version3.16.0 - 3.16.1
PlatformsWindowsmacOSLinux
I only have the Codex app locally, not the Codex CLI. Will I hit this problem too?
Yes. With only the Codex app installed you may have no ~/.codex/models_cache.json. Install the Codex CLI and generate the file with codex debug models --bundled.
What does running codex debug models --bundled do?
It prints the model list built into Codex and writes it into models_cache.json, correcting the model cache cc-switch reads when switching providers.
What if the Codex base_url changes after restarting cc-switch?
Maintainers say this overlaps heavily with the group of Codex configuration and model catalog refresh problems fixed in v3.16.0 / v3.16.1 — upgrade to v3.16.1 or newer and verify again.

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.