Run Claude Code on MiniMax-M3 and MiniMax-M2.7

Keep the Claude Code workflow, pay MiniMax prices: three environment variables connect Claude Code to MiniMax through an Anthropic-compatible endpoint.

Why run Claude Code on MiniMax

Claude Code talks to any server that speaks the Anthropic Messages API. YiduoChan exposes MiniMax-M3 and MiniMax-M2.7 through exactly that protocol, so you keep the Claude Code workflow, agents and tools while paying MiniMax prices: MiniMax-M2.7 costs $0.30 per million input tokens and $1.20 per million output tokens, and MiniMax-M3 gives you a 1,048,576-token context for whole-repository work at the same rate in the ≤512K tier.

Setup with environment variables

# 1. Get an API key at https://yiduochan.com/register
# 2. Point Claude Code at the Anthropic-compatible endpoint
export ANTHROPIC_BASE_URL=https://yiduochan.com
export ANTHROPIC_AUTH_TOKEN=YIDUOCHAN_API_KEY
export ANTHROPIC_MODEL=MiniMax-M2.7
export ANTHROPIC_DEFAULT_SONNET_MODEL=MiniMax-M2.7
export ANTHROPIC_DEFAULT_OPUS_MODEL=MiniMax-M3
export ANTHROPIC_DEFAULT_HAIKU_MODEL=MiniMax-M2.7-highspeed
# 3. Start Claude Code as usual
claude

Setup with settings.json

Put the same values in ~/.claude/settings.json (or the project's .claude/settings.json) for a persistent configuration:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://yiduochan.com",
    "ANTHROPIC_AUTH_TOKEN": "YIDUOCHAN_API_KEY",
    "ANTHROPIC_MODEL": "MiniMax-M3",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "MiniMax-M2.7",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "MiniMax-M3",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "MiniMax-M2.7-highspeed"
  }
}

Which model to choose

Prompt caching is automatic: repeated context is billed at the cache-read rate of $0.06 per million tokens, which matters for Claude Code's long, repetitive system prompts.

Two caveats. The ANTHROPIC_DEFAULT_*_MODEL variables map Claude Code's built-in sonnet, opus and haiku aliases to MiniMax models; the older ANTHROPIC_SMALL_FAST_MODEL is deprecated. The /v1/messages/count_tokens endpoint is not available on this relay yet, so Claude Code estimates context usage locally; regular requests are unaffected.

Verify the connection

curl https://yiduochan.com/v1/messages \
  -H "x-api-key: $YIDUOCHAN_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model": "MiniMax-M2.7", "max_tokens": 64, "messages": [{"role": "user", "content": "Say hello"}]}'

Getting a key without signing up

Creating an account and topping up is the usual route, and it is the one most people want. It is not the only one. The same models are exposed on an agent-native endpoint that speaks the x402 protocol: an unpaid request comes back 402 Payment Required with the terms in a header, and a request carrying a signed USDC authorization comes back with the completion, an API key and the remaining balance. No account, no email, no card.

You can read the terms without spending anything — the challenge is a plain JSON document, base64 in the PAYMENT-REQUIRED header:

curl -i -X POST https://yiduochan.com/api/x402/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "MiniMax-M2.7", "messages": [{"role": "user", "content": "hi"}]}'

It settles on Base and Arbitrum. This is the path to use when the caller is a program that has to decide at runtime which tools it needs, rather than a person who can go fill in a signup form. Details and the prepaid variant are on the agent guide.

Other Anthropic-compatible tools

• Cline / Roo Code: choose the Anthropic provider, set the base URL to https://yiduochan.com and enter the model name manually.
• Anthropic SDKs: construct the client with base_url="https://yiduochan.com" and your key.
• OpenAI-compatible tools such as Cursor or Continue use https://yiduochan.com/v1 instead; see the MiniMax-M2.7 guide.

FAQ

Sign up and get an API key See full model pricing