Meituan releases LongCat-2.5, a 1.6T-parameter multimodal agent model

Meituan released LongCat-2.5, a native multimodal mixture-of-experts model with 1.6 trillion total parameters and approximately 48 billion active parameters per token, optimized for long-horizon agentic coding and visual tasks. The model is available via API with a 1-million-token context window and competitive token pricing; open weights are stated as forthcoming. Company-reported results position it competitively on selected agent and browsing benchmarks against frontier closed models.