New model · rolling out now

    MiniMax M3 is coming to Kimchi Coding

    We're among the first platforms in the world to offer M3 access. It's rolling out in phases - join the waitlist to get it first.

    Why the builder matters

    When you hand a task to Kimchi Coding, the orchestrator classifies and delegates. Planning phases go to the planner. File reads go to the explorer. But the actual work - arguably the most important part - goes to the builder.

    It's where most of the tokens are spent and where model quality has the most impact. M3 replaces MiniMax M2.7 as the default builder, and it outperforms the previous model by a meaningful margin.

    59.0% on SWE-Bench Pro

    Benchmarked on real GitHub issues - ahead of GPT-5.5 and Gemini 3.1 Pro on the same test.

    1M token context, actually usable

    MiniMax Sparse Attention (MSA) makes long context fast - 1/20th the per-token compute of M2.7 at 1M length.

    15× faster decoding

    Builder subagents finish each /ferment step dramatically faster, and the gains compound across parallel agents.

    The new default builder

    Kimchi's orchestrator sends the heaviest work - writing the actual code - to the builder. M3 replaces M2.7 there.

    Get it first

    M3 is rolling out in phases via the waitlist. Drop your email and we'll let you know the moment it goes live.