arrow_backBack to Blog
AImodelsopen-sourceChinachipsgeopoliticsMeituan

Meituan Launches LongCat-2.0: First Trillion-Parameter AI Model Trained 100% on Chinese Chips

personLab Desenvolvimento
schedule5 min

Chinese delivery giant Meituan made a move that goes way beyond food: they released and open-sourced LongCat-2.0, claiming it’s the world’s first trillion-parameter language model trained and run entirely on a cluster of 50,000 chips made in China.

This is a direct challenge to US export restrictions, which aim to limit China’s access to cutting-edge AI chips (NVIDIA, AMD). The message is clear: you can do serious AI without American hardware.

The LongCat team, founded in 2023, entered China’s crowded AI sector late — dominated by DeepSeek and ByteDance Doubao. But with this release, they position themselves as proof that technological independence in AI is achievable, even if the road is harder.

The model was open-sourced, so you can actually play with it. An invitation to test how a Chinese-chip-trained model compares to the American giants.