Skip to content

FAQ ​

What should I use as the Base URL? ​

It depends on which level the tool or SDK expects:

Use https://api.codingrouter.ai/v1Use https://api.codingrouter.ai
OpenAI SDK, Codex CLI, OpenCode, most chat clientsAnthropic SDK, Claude Code, Google GenAI SDK, Gemini CLI

Without /v1, the request lands on the website home page and returns HTML. With an extra /v1, it usually returns 404.

Is streaming supported? ​

Yes. For the OpenAI and Claude APIs pass "stream": true. For the Gemini API call :streamGenerateContent?alt=sse. See the API reference.

How do I know whether a model is available? ​

Search for the model name in the model square. The details list its groups, price and supported APIs. You can also call List models with your token, which returns the models that token can use.

Can one token call every model? ​

No. A token is bound to one group and can only use the models in that group. To use several vendors, create a token for each.

The model name exists, so why do I get 503? ​

A 503 is also returned when the token's group does not contain the model. Check which group the model is in via the model square and use a token from that group.

Why does the charge differ from my calculation? ​

Common reasons:

  • The thinking process of reasoning models counts as output tokens.
  • Cached input hits are billed at the cache price.
  • Some models are tiered by context length, and the unit price changes above a threshold.

The details of every call are in the usage log.

A reasoning model returned an empty body? ​

Models such as GLM output their thinking first. If max_tokens is too small, the whole budget is spent on thinking, the body is empty and finish_reason is length. Raise max_tokens to 2000 or more.