Meituan's 1.6T-parameter LongCat-2.5 agent model trained without Nvidia GPUs
The Decoder · rss · 2026-11-15
Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter agent-focused model trained entirely on Chinese hardware with no mention of Nvidia GPUs. It scores 88.4 on Terminal Bench, just behind Claude Opus 5, and the sample deployment runs on 194 Huawei Ascend chips — a landmark case of frontier agent models trained on domestic compute.
More from Infra
- Team builds custom inference engine in 4 days for $500 in tokens, beats SGLang by 15% with 6 prompts — IgorCarron · 2026-11-15
- Google TPUv7 beats Nvidia B200/GB300 on inference cost in SemiAnalysis benchmarks — SumitGup · 2026-11-15
- Court Refuses to Dismiss DRAM Supply Conspiracy Suit Against Samsung, SK Hynix, Micron — SumitGup · 2026-11-15
- 2009 video shows Jensen Huang sizing CUDA at billions when NVDA traded at $0.2 — GregoryDiamos · 2026-11-15
- Ex-NVIDIA engineer recalls 2012: Jensen Huang backed the deep learning cluster behind Deep Speech — GregoryDiamos · 2026-11-15
- EU's Open Model EUROPA Eyes ~1T Params, Per-GPU Throughput Up 14x — mawcopolow · 2026-11-15