China's food delivery giant Meituan has stepped into the AI race, and the way it built its newest model is drawing as much attention as the model itself.
According to Reuters, Meituan said on Tuesday that it had released and would open-source LongCat-2.0, a 1.6-trillion-parameter model. The company says the system was trained on a cluster of 50,000 domestic Chinese processors, rather than the foreign chips that have long dominated cutting-edge AI work.
Reuters notes that Meituan did not provide details about the chips or the cluster. As Cybernews framed it, the company claims to have built the massive model using only domestic hardware.
A few things are worth keeping in mind. "Parameters" are a rough measure of a model's size and capacity, and at 1.6 trillion, LongCat-2.0 sits in the upper tier of publicly disclosed models. Open-sourcing means Meituan plans to make the model freely available for others to download and build on, a strategy several Chinese tech firms have embraced. And the 50,000-chip figure, if accurate, points to a training operation on a substantial scale.
Meituan is best known outside China as a food delivery and local-services platform, not an AI lab, which makes its entry into frontier model-building notable on its own.
The claims have not been independently verified, and the lack of technical detail leaves open questions about exactly which processors were used and how they performed.
Why it matters: with U.S. export controls limiting China's access to advanced foreign AI chips, a major Chinese company saying it trained a huge model on home-grown silicon is a signal about how far the country's push for technological self-sufficiency may be progressing.