For years, one of the biggest questions in artificial intelligence has been whether China could continue building cutting-edge AI systems without relying on American chips. Meituan believes the answer is yes.
The Chinese technology company announced the release of LongCat-2.0, a new open-source large language model that it says was trained and runs entirely on a cluster of 50,000 Chinese-made AI chips. According to the company, it is also the world’s first trillion-parameter AI model developed without foreign processors.
If those claims hold up, the announcement represents more than another AI model launch. It signals that China’s domestic AI ecosystem is becoming increasingly capable of developing advanced systems despite years of U.S. export restrictions on high-performance semiconductor technology.
More Than Just Another AI Model
Meituan is best known as one of China’s largest food delivery and local services platforms, often compared to DoorDash. Unlike companies such as DeepSeek, Alibaba, or ByteDance, it entered the generative AI race relatively late.
Its dedicated AI team was only established in 2023, and the first LongCat model arrived near the end of last year. LongCat-2.0 is therefore a significant leap forward in a remarkably short period of time.
The company has also confirmed that the model will be released as open source, allowing developers to inspect, use, and build upon its technology.
Why the Chips Matter
The most significant part of the announcement is not the model itself—it’s the hardware behind it.
Since 2022, the United States has imposed increasingly strict export controls that limit China’s access to advanced AI chips from companies like NVIDIA. Those restrictions forced Chinese technology firms to invest heavily in domestic alternatives.
According to Meituan, LongCat-2.0 was trained entirely on a computing cluster powered by Chinese processors, without relying on foreign AI hardware. Although the company did not identify the chip manufacturer, Chinese firms including Huawei and Enflame have rapidly expanded their AI chip production to meet growing domestic demand.
If independently confirmed, this would mark an important milestone in China’s effort to build a self-sufficient AI infrastructure.
Built for AI Agents and Software Development
Rather than positioning LongCat-2.0 as a general-purpose chatbot, Meituan says the model has been optimized for agentic coding—AI systems capable of completing complex software development tasks with minimal human intervention.
The model reportedly supports context windows of up to one million tokens, enabling it to process extremely large codebases and lengthy technical documents in a single session.
Meituan also highlighted practical demonstrations showing LongCat-2.0 building a gaming website and generating long-form creative writing, illustrating the broader capabilities of the system beyond programming.
AI Is Becoming Part of Meituan’s Business
While the company has not detailed how LongCat-2.0 will be integrated into its products, earlier versions of the model already power AI assistants inside Meituan’s apps.
These assistants help users discover restaurants and hotels, recommend services, place food delivery orders, and complete travel bookings. The strategy reflects the industry’s growing shift toward AI agents capable of performing tasks instead of simply answering questions.
For Meituan, AI could also become an important new source of growth as competition intensifies and margins in its core delivery business continue to shrink.
Bold Performance Claims
Meituan says a preview version of LongCat-2.0 quickly became one of the three most-used models on OpenRouter, a popular marketplace where developers compare and access different AI models.
The company also claims that LongCat-2.0 matched or outperformed several leading proprietary models—including OpenAI‘s GPT-5.5, Google’s Gemini, and Anthropic’s Claude Opus—across selected coding and AI agent benchmarks.
These results have not yet been independently verified, so they should be viewed as company-reported performance rather than established industry consensus.
Why This Matters
The bigger story isn’t simply that another company released another AI model.
China is steadily proving that it can continue advancing its AI capabilities despite losing access to many of the world’s most advanced American chips. Instead of slowing down development, export restrictions appear to be accelerating investment in domestic processors, local AI infrastructure, and open-source innovation.
Whether LongCat-2.0 ultimately lives up to its technical claims remains to be seen. But its launch sends a clear message: the global AI race is no longer defined only by who builds the smartest model, but also by who controls the hardware needed to train the next generation of AI.
