AI & Tech··1 sources verified
Meituan Open-Sources LongCat-2.5, a 1.6T-Parameter Agent Model Under MIT License
Summarised by Relevant News AI · Read time: 3 min
Meituan released LongCat-2.5 weights under the MIT license, with 1.6 trillion total parameters, about 48B active per token, and text, image, and video input. Meituan's self-reported results put it at 88.4 on Terminal Bench v2.1, near Claude Opus 5's 89.1, but at 38.5 versus 53.9 on the harder Terminal Bench v4.0. The model card documents only Huawei NPU deployment, with no Nvidia path yet, and the API reportedly supports both OpenAI and Anthropic formats.
Why it matters: A large open-weights agent model with MIT terms and Anthropic- and OpenAI-compatible APIs could be cheap to test in existing agent pipelines, though the benchmark gap and NPU-only deployment guidance mean its real-world coding performance is still unverified.
All sources