Meituan Trains LongCat-2.0 on Domestic 50,000-Chip Cluster

Meituan released and open-sourced LongCat-2.0, a 1.6-trillion-parameter large language model with a 1-million-token context window, trained from scratch on a 50,000-chip domestic compute cluster. The company claims it is the industry's first trillion-parameter model to complete end-to-end training and inference on Chinese-made processors, though it did not disclose the chip supplier.

Industry context: Demonstrations of end-to-end training on home-grown hardware affect how practitioners assess supply-chain risk, performance trade-offs, and reproducibility. Reuters reports Meituan released and open-sourced a new large language model, LongCat-2.0, which multiple outlets report carries 1.6 trillion parameters and a 1 million-token context window. The Next Web quotes Meituan describing LongCat-2.0 as "the industry's first trillion-parameter model to complete end-to-end training and inference on a 50,000-chip domestic compute cluster." Reuters and SCMP report the model was trained from scratch on a 50,000-chip cluster powered by Chinese-made processors and that Meituan published the weights. CNA and SCMP note Meituan did not disclose which domestic chipmaker supplied the hardware.