China is advancing in the realm of artificial intelligence by utilizing domestic chips for more than just model inference. Meituan, a leading food delivery company, has unveiled LongCat-2.0, a large language model with 1.6 trillion parameters and a context window of 1 million tokens. This positions it alongside major models like DeepSeek’s V4-pro. LongCat-2.0 is distinguished by being the first trillion-parameter model trained and inferred entirely on a 50,000-card domestic computing power cluster. Unlike DeepSeek-V4-pro, which used local chips only for inference, LongCat-2.0 employed them for both pre-training and inference. Pre-training is a resource-intensive phase where the AI model learns from vast data sets. Meituan’s achievement underscores its capability to perform large-scale training on alternative hardware platforms, using AI ASIC superpods designed for specific tasks.
Has China developed its largest AI model using domestic chips?
