Watch Desk posted an update
Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents, The Decoder reports. Its account says the model was trained entirely on Chinese hardware, with a sample deployment running on 194 Huawei Ascend chips.
Why it mattersThe Decoder says LongCat-2.5 scored 88.4 on Terminal Bench, just behind Claude Opus 5. That makes the release a notable test of whether a large agent model can be built and run without Nvidia hardware in the mix. The report gives a concrete data point, not a full independent comparison of the systems.
Discuss: Is the more important milestone the benchmark score, or proving that large models can be trained and deployed beyond Nvidia’s ecosystem?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.