China's food delivery giant Meituan has released a new version of its LongCat large language model and says it was trained on domestic chips rather than foreign hardware.
According to Reuters, whose report was bylined by Ethan Wang and Eduardo Baptista from Beijing, Meituan announced on Tuesday, June 30, that it had released its next-generation LongCat model and would make it open-source — meaning other developers can freely download, study and build on it.
The South China Morning Post described the release as China's biggest AI model trained on local chips. Crypto Briefing identified the system as LongCat-2.0, a 1.6-trillion-parameter coding model, and said it was trained entirely on Chinese chips. (Parameters are a rough measure of an AI model's size and capacity; more of them generally signals a more powerful system.)
The story was carried widely across Asian and international outlets, including CNA, TradingView, The Edge Malaysia and The Standard in Hong Kong, reflecting the level of attention paid to any sign of progress in China's homegrown chip and AI efforts.
What makes the claim notable is the hardware. Cutting-edge AI is usually trained on advanced processors from companies such as Nvidia, and US export controls have restricted Chinese firms' access to the most powerful versions. A major Chinese company saying it built a large, capable model without relying on top foreign chips suggests the country is finding workarounds.
Why it matters: if Chinese firms can train frontier-scale AI models on domestic chips, it weakens the leverage of US export restrictions and signals that China's push for technological self-reliance is gaining ground.