Release
Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU
The Decoder · By Jonathan Kemper ·
The Brief
- Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents.
- The model was reported as trained entirely on Chinese hardware, with no Nvidia GPUs used.
- It scores 88.4 on Terminal Bench, just behind Claude Opus 5, and a sample deployment runs on 194 Huawei Ascend chips.
Why it matters
It shows a very large agent-oriented model can be trained and deployed on non-Nvidia hardware, which matters for teams weighing hardware options and open-weight alternatives for agent workloads.
Who’s affected
- Teams evaluating open-weight agent models
- Organisations considering Huawei Ascend or other non-Nvidia AI hardware
- #ai-models
- #agents
- #open-weights
- #huawei-ascend
- #benchmarks
- #meituan
Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents and trained entirely on Chinese hardware. On Terminal Bench, it scores 88.4 points, landing just behind Claude Opus 5. The sample deployment runs on 194 Huawei Ascend chips, and…