Meituan's LongCat Model Leads Global AI Rankings with Breakthrough Scale
According to data from OpenRouter and reported by Techub News, Meituan's LongCat-2.0-Preview model has surged to the top of global AI rankings, processing an unprecedented 10.1 trillion tokens per month. The model boasts 1.6 trillion parameters and was trained entirely on a cluster of 50,000 to 60,000 domestic Chinese AI accelerators, with zero reliance on NVIDIA hardware. This achievement marks a major milestone for Chinese AI labs operating outside the NVIDIA ecosystem.
Launched for public beta on April 24, 2026, LongCat-2.0-Preview supports a massive 100 million token context window and includes built-in tool calling capabilities. The training infrastructure relied solely on chips from Chinese vendors such as Huawei's Ascend series and Cambricon, validating the feasibility of scaling up large model training on domestic hardware. This reduces dependency on foreign supply chains and opens new pathways for high-performance AI compute in China.
Implications for the Crypto Industry: Decentralized Compute and Chip Independence
The LongCat model's technical success carries direct relevance for the crypto sector, particularly in decentralized AI (DeAI) and decentralized physical infrastructure networks (DePIN). Projects like Render Network, Akash Network, and io.net are aggregating idle global GPU power to serve AI training workloads. Meituan's demonstration that a top-tier model can be trained on non-NVIDIA chips broadens the hardware landscape for crypto-based compute markets, potentially attracting more participants and reducing cost barriers.
Amid ongoing US-China technology decoupling, the intersection of AI and blockchain increasingly demands chip self-sufficiency. The LongCat breakthrough may incentivize Web3 projects to adapt their software stacks for domestic chips, accelerating DePIN adoption in Asia. If domestic chip manufacturers open compute trading interfaces, they could become major supply nodes in decentralized compute markets, further lowering AI training costs and enhancing privacy through on-chain execution.
While Meituan has not disclosed any plans to integrate blockchain tokens or incentives into LongCat, the model's massive scale and domestic origin have sparked renewed discussions in crypto communities about the feasibility of on-chain AI. As public beta data emerges, the model's influence on AI+Web3 convergence will become clearer.