On June 30, Meituan introduced LongCat-2.0, a new large-language (LLM) model that the company claims has performance equivalent to Gemini 3.1 Pro released by Google in February.
Large language models are the technology platform behind chatbots and many other AI tools. The more parameters the model has, the higher the ability to process complex requests.
According to Meituan, LongCat-2.0 is the industry's first AI model with over 1,000 billion fully trained parameters and operating on a cluster of 50,000 chips manufactured by China.
However, the company did not disclose which Chinese manufacturer provided the chips used in the training process.
This announcement is considered an important milestone for China's AI industry because training competitive AI models with huge amounts of data requires processors with very strong computing capabilities.
Key AI models of other AI research labs in China, including DeepSeek and Zhipu, can now perform the inference stage on chips manufactured by domestic enterprises such as Huawei. However, inference is a process that consumes much less computing resources than model training.
Meituan said that the company's AI research team began testing the use of domestic chips in 2023 and has now proven that it is possible to train large-scale AI models on domestic computer clusters.
Meituan has also invested in many domestic semiconductor enterprises in China, including Moore Threads and MetaX. Shares of both companies have increased sharply after listing on the stock exchange in recent months.
