News
MiniMax M2.5 Is Out & It's Benchmark Scores Are Shocking! It's Open-Source & It's Better Than Most!
calendar_today Date:
schedule Duration: 0:56
visibility Views: 773
database
Summary Report
MiniMax M2.5 is out, and the gap between a state of the art model from one of the big four and an equivalent open-source model like this is now, basically zero. MiniMax has announced the M2.5 model,
MiniMax M2.5 is out, and the gap between a state of the art model from one of the big four and an equivalent open-source model like this is now, basically zero.
MiniMax has announced the M2.5 model, an open-source LLM from Shanghai. So what kind of performance fall off do you see if you're just taking an off the shelf open-source model like this? None. It matches the coding performance of the best models in the world like Claude Opus 4.6, and smashes Gemini and GPT 5.2 across many benchmarks. It's also said to be 37% faster than prior MiniMax versions. If you're running some crazy 24 hour a day Molty then this is probably the model for you. [thoughtful] Just imagine the pressure this is putting of the likes of Google and OpenAI.