🇨🇳China claims a new milestone in locally trained AI, as Meituan rolls out LongCat-2.0.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-30
Meituan, China's food delivery giant, releases LongCat-2.0, an open-source 1.6T-parameter MoE coding model trained entirely on 50,000 Chinese domestic chips, marking the first large-scale Chinese LLM trained on domestic hardware for both pre-training and inference.
Appears in
Extraction
Topics: chinese-aidomestic-chipsopen-source-llmexport-controlsmixture-of-experts
Claims
- Meituan released LongCat-2.0, a 1.6T-parameter MoE model with a 1M token context window trained from scratch on 50,000 Chinese domestic chips.
- LongCat-2.0 used domestic hardware for both pre-training and inference, going further than DeepSeek-V4-pro which only used home-grown chips for inference.
- Meituan used Huawei's HCCL chip-to-chip communication library to stabilize training, implicating Huawei Atlas-950 hardware as the compute base.
- Chinese AI companies are accelerating domestic chip self-reliance following U.S. export controls enacted since 2022.
- LongCat-2.0 ranks in the top 3 globally by call volume on OpenRouter.
Key quotes
LongCat-2.0 was trained from scratch on 50,000 Chinese domestic chips and Meituan said this proves large-scale model training can now be done on domestic compute clusters.
While DeepSeek-V4-pro relied on home-grown chips only for inference, LongCat-2.0 used domestic hardware for both inference and pre-training, according to Meituan.
This removed doubts that Atlas-950 SuperPoDs could not train large LLMs for Zhipu AI and DeepSeek.