China’s AI industry just got another major player.
Meituan, the company best known for food delivery and local services in China, has officially unveiled LongCat 2.0, its most advanced AI model yet. While Meituan isn’t usually mentioned alongside companies like OpenAI or Google, this release has quickly caught the attention of the AI community.
The reason isn’t just the model itself.
It’s how it was built.
According to Meituan, LongCat 2.0 is the first trillion-parameter AI model to be trained entirely on Chinese-made AI chips instead of NVIDIA hardware. If that claim holds up, it marks a significant milestone for China’s AI industry as it continues to develop its own AI infrastructure despite U.S. export restrictions.
What is LongCat 2.0?
LongCat 2.0 is Meituan’s newest open-source large language model. The company says it’s designed for demanding tasks like coding, reasoning, AI agents, and analyzing extremely long documents.
One of its standout features is a 1 million token context window. In simple terms, that means the model can process enormous amounts of information in a single conversation. Entire books, large codebases, or lengthy research papers can fit into one prompt without needing to split them into smaller pieces.
A huge focus on AI agents
Meituan isn’t positioning LongCat 2.0 as just another chatbot.
Instead, the company says it was built specifically for agentic AI—AI systems that can complete multi-step tasks with minimal human guidance.
That includes things like writing software, creating websites, analyzing documents, and carrying out complex workflows that require planning and reasoning instead of simply answering questions. The company says the model performs especially well in coding and long-running AI agent tasks.
As AI agents become one of the biggest trends in the industry, that’s a smart direction to focus on.
Trained without NVIDIA chips
Perhaps the biggest headline has nothing to do with benchmarks.
Meituan says LongCat 2.0 was trained on a cluster of around 50,000 Chinese-made AI chips, making it the first frontier-scale AI model to complete both training and inference entirely on domestic hardware.
For years, many experts believed China’s biggest obstacle in AI would be limited access to advanced NVIDIA chips.
LongCat 2.0 suggests Chinese companies are finding ways to build powerful AI systems using their own hardware instead. That’s a major development in the global AI race.
How does it compare to other AI models?
Meituan claims LongCat 2.0 can compete with some of today’s leading frontier AI models in areas like reasoning, coding, and long-context tasks.
The company also says the model has become one of the top-performing open models available through AI marketplaces shortly after launch. However, these are Meituan’s own performance claims, and broader independent testing is still ongoing.
As more developers begin testing the model, we’ll get a much clearer picture of where it stands against competitors like GPT-5.5, Claude, Gemini, DeepSeek, and Qwen.
Why this matters
LongCat 2.0 is important for more than just its technical specifications.
It shows how quickly China’s AI ecosystem is evolving.
Companies that were once focused on food delivery, e-commerce, or social media are now building some of the world’s largest AI models. At the same time, many of these models are being released as open source, allowing developers around the world to experiment with them and build new applications.
That creates more competition, more innovation, and more choices for developers.
Final thoughts
LongCat 2.0 is one of the most surprising AI releases of the year.
Not only is it a massive open-source model built for coding and AI agents, but it’s also a demonstration that China can train frontier-scale AI systems using domestic hardware.
There’s still plenty of independent testing to be done, so it’s too early to declare it the new leader of the AI world. But one thing is already clear: Meituan has gone from being known as a food delivery giant to becoming a serious competitor in the global AI race.
As more developers get their hands on LongCat 2.0, we’ll soon find out whether it can truly live up to the excitement.