Using Any Other LLM Inside Claude Code

Using Any Other LLM Inside Claude Code

Plug and switch any model you want to use in claude code for flexibility and cheaper option.

ai dev
Table of Contents

Currently there is a lot of people are using Claude Code, terminal-based ai tool to assist their programming or to vibecode. The ability and flexibility to do bunch of tasks done through CLI and tool enabled by CC is really powerful and useful personally. Sadly, CC is dedicated for anthropic models only such as Sonnet and Opus. So, here is the “tricks” I have been using to switch and use any models beside anthropic models especially OSS models that have comparable performance to Sonnet itself (ex: Qwen3 Coder, Kimi K2, GLM 4.5). Thus, you can use cheaper model you want inside claude code directly.



Using Proxy Server

The first method that I have been using in the past is running proxy server that enables CC to work with any Open-AI compatible API providers. Basically you map any other LLM endpoint and its env cofiguration into Anthropic comparable client then you run localhost server to plug it into Anthropic client configuration such as ANTHROPIC_BASE_URL=http://localhost:8082 #localhost FastAPI server inside CC.

TLDR: convert OpenAI API–compatible requests into Anthropic API format > Set the environment variables > override CC baseUrl and apiKey

There is already several claude code proxy github repos that you can directly use so you dont have to implement it from scratch:



Configure The Environment

Well you know the idea, as basically we just set and inject other LLM env configuration into CC baseUrl & apiKey, we can simplified the setup more, no need to start localhost server as proxy just to use other LLM in CC. Many AI labs such as Moonshot AI and Z.AI already make CC-compatible API endpoint that you can inject directly into Anthropic base URL. Now, Z.AI provides claude code plan for $3/month and $15/month be to be used specifically inside claude code (which is a lot of cheaper than claude code max plan). Thus, you can export the model baseUrl and apiKey directly then run claude like this:

export ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
claude

Also you can configure the model on ~/.claude/settings.json with:

{
  "env": {
      "ANTHROPIC_MODEL": "glm-4.5"
  }
}

After that, you can run claude then prompt /status. It should be glm-4.5 on there. As we know that we only need to export the configuration directly then run claude to switch model now, we can wrap it as bash function and assign it as glm or whatever your model name is. Now, you can just run glm for running Z.AI model inside CC.

glm() {
    export ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic
    export ANTHROPIC_AUTH_TOKEN=$ZAI_API_KEY
    export ANTHROPIC_MODEL="glm-4.5"
    claude $1
}

You can also do this to other model that provide CC-compatible baseUrl like Moonshot one.

kimi() {
    export ANTHROPIC_BASE_URL=https://api.moonshot.ai/anthropic
    export ANTHROPIC_AUTH_TOKEN=$MOONSHOT_API_KEY
    export ANTHROPIC_MODEL="kimi-k2-0905-preview"
    claude $1
}


References