Key facts
- Meituan released its next-generation LongCat-2.0 large language model.
- The model is claimed to be the world's first trillion-parameter AI system trained and run entirely on domestic Chinese chips.
- The training utilized a 50,000-chip cluster powered by Chinese-made processors.
- LongCat-2.0 can process inputs of up to 1 million tokens, enabling it to handle ultra-long documents.
- Meituan stated the model is aimed at agentic coding and demonstrated capabilities in building websites and writing novels.
China's food delivery giant Meituan announced on Tuesday the release and open-sourcing of its LongCat-2.0 large language model. The company claims it is the world's first trillion-parameter AI system trained and run entirely on a cluster of 50,000 domestic Chinese chips.
This development underscores China's drive for self-sufficiency in AI, as major players like Meituan, DeepSeek, Alibaba, and ByteDance aim to reduce reliance on U.S. chips due to Washington's export controls imposed since 2022. Chinese chipmakers such as Huawei and Enflame are reportedly stepping in to meet this demand.
The LongCat-2.0 model is designed for agentic coding and can process extensive inputs of up to 1 million tokens. Meituan stated that a preview version performed competitively against leading proprietary models like Google's Gemini, OpenAI's GPT-5.5, and Anthropic's Claude Opus on certain benchmarks. The company also highlighted the model's creative capabilities, including building a gaming website and writing a novel.