The Decoder· Jonathan Kemper·· 13 h ago
Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU
Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU
AI summary
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 Meituan makes no mention of Nvidia GPUs.
Selection record
Threshold 76Media and individualsFirst 88Second 88
AdmittedSum of both 176 ≥ twice the threshold 152
- Source tier
- Media and individuals; this tier's threshold is 76
- Pre-filter
- passed：美团发布1.6万亿参数Agent模型，明确AI技术
- Why it was chosen
- Meituan has open-sourced a 1.6-trillion-parameter agent model trained and deployed entirely on Chinese chips, offering a test case for how far domestic compute can carry large models.
A model scores each item twice, independently, against one written standard, out of 100. An item is admitted only when the two scores add up to twice the threshold. The threshold is set per source tier.
Source: The Decoder · the-decoder.com