Meituan's LongCat achieves 1.6 trillion parameter model training entirely on Chinese chips, with no Nvidia required.

Meituan has trained a 1.6 trillion parameter AI model entirely on domestically manufactured chips, demonstrating what the company says is a major milestone in Chinese AI development independent of Nvidia technology. The training ran on a cluster of more than 50,000 domestically made AI ASICs and processed over 35 trillion tokens. The LongCat team, which began work in 2023, shipped its first model late last year.

"LongCat-2.0 has demonstrated that we now have the capability to train large-scale models on domestic computing clusters," Meituan said in announcing the breakthrough.

Performance on standardized benchmarks is mixed. The model leads on some tests: it scores 59.5 on SWE-bench Pro and 77.3 on SWE-bench Multilingual, outperforming Gemini 3.1 Pro and GPT-5.5. However, it falls short of Claude Opus 4.7 and 4.8 on these measures. On other benchmarks—IFEval (90.0), IMO-AnswerBench (81.8), and GPQA-diamond (88.9)—it trails Gemini and GPT-5.5 by significant margins in several cases.

The implications for US-China competition in AI are significant. Despite American export controls on advanced semiconductors in place since 2022, China appears to have produced what may be its first competitive trillion-parameter model trained entirely on domestic hardware. The effort remains somewhat opaque: Meituan has not disclosed the specific chip manufacturer, and the model is not yet available on HuggingFace, limiting independent verification of the claims.