
LongCat-2.5
Launching todayMeituan's 1.6T open-weight model for long agent tasks
LongCat-2.5 is Meituan's open-weight model for long agentic work: coding across whole codebases and full projects, plus visual tasks such as design and 3D. It is a mixture-of-experts model with 1.6 trillion parameters, about 48 billion active per token, native image input and a 512K context. Meituan says it was trained entirely on Chinese-made chips; its benchmark scores are self-reported. The weights are MIT-licensed, and you can try it in the chat at longcat.ai.
Meet the team
Hunted by Ilyas. The makers are not on 25launches yet.


