Codex sends prompt_cache_retention to models that don't support it, causes 400

[skrabe]

gpt-5.3-codex requests fail with 400 because CLIProxyAPIPlus sends prompt_cache_retention=24h and OpenAI rejects it.

Debug log:

request error, error status: 400, error message: {"detail":"Unsupported parameter: prompt_cache_retention"}

OpenAI's docs only list these models as supporting extended prompt cache retention: gpt-5.4, gpt-5.2, gpt-5.1, gpt-5.1-codex, gpt-5.1-codex-mini, gpt-5.1-codex-max, gpt-5, gpt-5-codex, gpt-4.1

gpt-5.3-codex is not on that list. The parameter also might not work through OAuth endpoints at all (only direct API).

Workaround via config:

payload:
  filter:
    - models:
        - name: "gpt-5.3-codex*"
      params:
        - "prompt_cache_retention"

Fix should check the model against the supported list before injecting this parameter.

Version: 6.9.5-0

[skrabe]

Update: this affects all GPT models through OAuth, not just gpt-5.3-codex. gpt-5.4 also gets rejected. The extended cache retention parameter likely only works through direct API keys, not OAuth/subscription endpoints. Updated workaround to filter all gpt-* models.