Meituan open-sources trillion-parameter AI model trained on Chinese chips

Chinese food delivery giant Meituan on Tuesday released and open-sourced LongCat-2.0, a 1.6 trillion-parameter large language model trained entirely on domestically produced chips — what the company claims is the world's first trillion-parameter AI system built and run on a cluster of 50,000 Chinese-made processors.money.usnews+1

The Mixture-of-Experts model, released under the permissive MIT License on GitHub and Hugging Face, features a one-million-token context window and is designed for autonomous coding, software engineering, and AI agent tasks. Meituan said LongCat-2.0's performance is comparable to Google's Alphabet Inc. ↘ 0.48% Gemini 3.1 Pro and matched or exceeded several leading proprietary models including OpenAI's GPT-5.5 and Anthropic's Claude Opus on key benchmarks.channelnewsasia+2

A Milestone for China's AI Self-Sufficiency

The release marks a notable advance in China's efforts to develop frontier AI without relying on Nvidia ↘ 1.39% hardware. While DeepSeek's V4-pro, launched in April, used domestic chips only for inference — the process of running a trained model — LongCat-2.0 used Chinese ASIC chips for both pre-training and inference, according to the South China Morning Post.scmp

"LongCat-2.0 has demonstrated that we now have the capability to train large-scale models on domestic computing clusters," Meituan said in a statement, without naming the chipmaker.wtaq+1

Meituan's AI research team began exploring the use of domestic chips in 2023. The model was previously known anonymously as "Owl Alpha" and had topped OpenRouter developer rankings before its identity was revealed.techxplore+2

Scale and Competition

At 1.6 trillion parameters, LongCat-2.0 puts Meituan on par with DeepSeek's latest flagship V4-pro model. The scale represents a leap from the company's earlier LongCat-Flash, which had 560 billion parameters.medium.datadriveninvestor+1

The model was built on what Meituan described as "large-scale clusters of tens of thousands of AI ASIC superpods," demonstrating its ability to "conduct frontier-scale training on alternative hardware platforms". The open-source release under an MIT License makes it one of the most commercially flexible large-scale AI models available, allowing developers to modify and redistribute it without restriction.scmp+2