Meituan open-sources LongCat-2.0, a 1.6T-parameter model that it says was trained on a 50K-chip cluster of domestic Chinese processors, without giving details
China's food delivery giant Meituan (3690.HK) said on Tuesday it had released and would open-source its next-generation LongCat …
Lead Source
More
Discussion
TechSnif Coverage
Meituan Drops LongCat-2.0: 1.6T Parameters, All-Chinese Chips
China's food delivery giant open-sources a massive AI model trained entirely on domestic processors.
Meituan just open-sourced LongCat-2.0, a 1.6 trillion-parameter model that makes a bold statement about Chinese AI independence. The food delivery titan says it trained the model on a cluster of 50,000 domestic Chinese processors.
Notable: Meituan didn't specify which chips it used. No vendor name, no architecture details. Just "domestic Chinese processors." That's a deliberate flex — and a deliberate omission.
The scale here is significant. A 50K-chip training cluster is serious infrastructure, and doing it without relying on Western silicon (at least, that's the claim) signals China's AI hardware ecosystem is maturing fast.
Meituan is better known for delivering takeout than pushing frontier AI. LongCat-2.0 suggests the company has bigger ambitions. Open-sourcing a model this size also throws down the gauntlet to other Chinese tech giants racing to build competitive large language models.