Meituan Releases Trillion-Dollar Model for Domestic Chips (LongCat)

According to Bijie.com, Meituan officially released its first trillion-parameter large-scale model, LongCat-2.0, which uses domestically produced chips for both training and inference. This model boasts 1.6 trillion total parameters, with approximately 48 billion parameters activated per token, and supports ultra-long contexts of up to 1 megabytes. LongCat-2.0 completed pre-training on 35 trillion tokens on a cluster of over 50,000 domestically produced AI chips, successfully verifying the engineering stability of domestic computing power supporting cutting-edge large-scale models. Furthermore, LongCat-2.0 was announced as fully open-source, licensed under the MIT license, releasing all weights and complete GPU and NPU inference deployment code.