Tag
Meituan released LongCat-2.0, a 1.6T-parameter MoE model with 1M context, claimed as the first to train on a 50,000-GPU Chinese cluster, now available on OpenRouter for agentic coding.