Meituan launches LongCat-2.0, first trillion‑parameter AI model trained entirely on Chinese chips

China’s food‑delivery giant Meituan announced the release of LongCat‑2.0, a 1.6‑trillion‑parameter mixture‑of‑experts large language model that it claims is the world’s first trillion‑parameter system trained and run solely on a cluster of about 50,000 domestically produced AI chips. The model was trained from scratch on Chinese ASICs, processing up to one million tokens and targeting agentic coding and commerce tasks. Meituan said LongCat‑2.0 matches or exceeds leading models such as Google Gemini, OpenAI’s GPT‑5.5 and Anthropic’s Claude Opus on certain benchmarks, and it has already become one of the most‑used models on the OpenRouter marketplace. The company is open‑sourcing the model and offering API access at $0.75 per million input tokens and $2.95 per million output tokens, with promotional pricing lower. The launch underscores China’s push for AI self‑sufficiency amid U.S. export controls on advanced chips, positioning Meituan alongside domestic rivals like ByteDance and DeepSeek in the competitive AI landscape.