Meituan has launched its LongCat-2.0-Preview AI model for testing, boasting over one trillion parameters, placing it among the world's leading large models. This development marks a significant milestone as the entire training and inference process was conducted using domestic computing infrastructure. Meituan utilized between 50,000 and 60,000 computing chips, setting a record for the largest model training task on domestic hardware. On the same day, DeepSeek announced its V4 large model, which reportedly has similar parameter counts to Meituan's LongCat-2.0-Preview. This highlights the competitive landscape in AI model development, with both companies pushing the boundaries of computational capabilities.