model · MiniMax

MiniMax-M2

MiniMax-M2 is a sparse mixture-of-experts model (230B total, only 10B active per token) built for coding and agentic work — multi-file edits, code-run-fix loops, and long tool chains. At ~205K context and ~115–130 GB in 4-bit, it actually fits on a 128 GB unified-memory box or a small multi-GPU rig, so you can genuinely self-host it — and its 'modified MIT' licence permits commercial use. It is also very cheap via the MiniMax API. Note the newer MiniMax-M2.7 is stronger but ships under a non-commercial licence.

• paid
• self-host
• Commercial