Comment by zeroxfe
That tends to work quite poorly because Claude Code does not use standard completions APIs. I tried it with Kimi, using litellm[proxy], and it failed in too many places.
That tends to work quite poorly because Claude Code does not use standard completions APIs. I tried it with Kimi, using litellm[proxy], and it failed in too many places.
It worked very well for me using qwen3 coder behind a litellm. Most other models just fail in weird ways though.
opencode is a good alternative that doesnt flake out in this way.
You can try Kimi's Anthropic-compatible API.
Just connect Claude Code to Kimi's API endpoint and everything works well
https://www.kimi.com/code/docs/en/more/third-party-agents.ht...