How to Use the CC Switch?

Sea!
1

What does it mean when they say a non-response model can’t handle codex? What is model mapping?

c8cbf9a50b0b4826c328554a8a8b8ccf0a4db8a9eff3584bb44f553f0daeb3a5
c8cbf9a50b0b4826c328554a8a8b8ccf0a4db8a9eff3584bb44f553f0daeb3a5710×832 55.7 KB

2

I haven’t used CC Switch before, but most of the terminology is universal, so I’ll give it a try to answer.

  1. Non-response model cannot receive codex
    This means that if the model API connected does not support the response format, it cannot be used in Codex. The conventional API path is https://api.xxx.com/v1, under which, the path of APIs that support response is /v1/responses, and the path of APIs that do not support response is /v1/chat/completions. The former can be used in Codex, while the latter cannot, because Codex requires APIs to be in response format by default.
  2. Model Mapping
    CC Switch’s explanation is

Some third-party providers only support OpenAI Chat Completion Protocol or use non-GPT model names (such as Baidu Qianfan, StepFun, or SiliconFlow). Codex itself only understands OpenAI Responses API and GPT series models, so these providers need CC Switch to convert protocols and models locally.

It can be simply understood that the program sends a request to CC Switch, the model ID is gpt-5.6, and after configuring model mapping, CC Switch will replace gpt-5.6 with other values before sending it to the model API. Model mapping can also be used to solve the problem that providers only support Chat Completions protocol (document explains this).