MiniMax
A World-Leading General AI Technology Company
189 followers
A World-Leading General AI Technology Company
189 followers
Founded in early 2022, MiniMax is driven by the mission to 'co-create intelligence with everyone,' dedicated to advancing the frontiers of AI and achieving Artificial General Intelligence (AGI). MiniMax has independently developed a series of multimodal foundation models with powerful code and Agent capabilities, as well as ultra-long context processing, capable of understanding, generating, and integrating multiple modalities including text, audio, image, video, and music.
Humans in the Loop
Big news for open models: MiniMax-M2.5 is out with SOTA performance at coding (SWE-Bench Verified 80.2%). The first open model to beat Sonnet. Only @Claude by Anthropic's Opus and @OpenAI 's GPT-5.2 Codex score higher.
Paths between open and proprietary models are converging...
Pro tip: If you want to quickly experiment with it, @MiniMax-M2.5 is free for a week on @Kilo Code (until Thursday, Feb 19).
OSS ftw!
@fmerian Whoa, MiniMax-M2.5 dropping SOTA SWE-Bench scores at 80.2% and beating Sonnet? Probably first open model to do it! 😲 Good hunt, @fmerian! :)
Humans in the Loop
@rohanrecommends yes! @MiniMax-M2.5 isn't designed to be the biggest - it's designed to scale in real production environments.
80%+ on SWE-Bench Verified for an open model is wild — especially if it’s actually usable in real workflows and not just benchmark-flexing. Curious how it holds up on messy, legacy codebases vs clean benchmark repos?
@andrius_aguinaldo SWE-Bench repos are clean, well-tested, and isolated. Real codebases have implicit conventions, undocumented side effects, and tests that pass for the wrong reasons. MiniMax M2 and M2.1 had reports of reward-hacking where the model changed existing tests to make its code pass instead of actually fixing the issue. M2.5 at 80.2% is promising but the 230B MoE architecture only activating 10B params per pass means you get the cost win at the expense of deep cross-file reasoning. Worth running it on a real refactor before trusting the number.
vibecoder.date
Awesome!
is it available for opencode yet?
Humans in the Loop
apparently! see pricing: https://opencode.ai/docs/zen/#pricing
looks great! This is something that seems like it would pair well with ClawdBot agents...
Vela
Congrats folks good luck on the launch!
Super Comments
Love it, I think i will add it to my saas YouScaleIt
That SWE-Bench score is wild for an open model. I've been running Sonnet for most of my coding workflows and honestly the cost adds up fast when you're doing long agentic runs. $1/hr with 100 tps would be a game changer if the quality holds up in practice. Curious - how does it handle multi-file refactors? That's where I see most models fall apart, they lose context across files even when the benchmarks look great.