MiniMax M3 API: pricing, base URL and quick setup

Specifications and pricing

Prices synced with platform billing (USD per 1M tokens).

Quick setup

curl https://a2agent.me/v1/chat/completions \
  -H "content-type: application/json" \
  -H "authorization: Bearer sk-your-a2agent-key" \
  -d '{
    "model": "minimax-m3",
    "messages": [{"role": "user", "content": "Reply with OK"}]
  }'

Using it from a coding client

Suits agent frameworks that plan, call tools and re-read state on every turn. The 1M window absorbs long transcripts, and the flat rate keeps long runs predictable.

• Run it in Claude Code
• Run it in Codex CLI

How it compares within its family

M3 is the agent-tagged MiniMax model and pairs a 1M context with the same $0.30 per million input tokens and $1.20 output that M2.5 and M2.7 list. Price never moves inside this family, so the decision comes down to what each model was tuned for and how much it can hold. M3 and M2.5 both reach 1M; M2.7 stops at 200K. M2.5 and M2.7 carry the coding tag, M3 carries agent. The practical split: tool loops, planning and orchestration on M3, writing and refactoring code on M2.5, and M2.7 when you want the newest release and the shorter window is not a problem. Switching between them costs nothing but the model string.

Other models on this platform

• MiniMax M2.51M
• MiniMax M2.7200K

Questions