China's First Self-Developed Trillion-Parameter Large Model for Card Training and Inference Officially Open Source - 36 Kr

Oh my goodness, this is the most GENEROUS gift to the global AI community I've ever seen! ✨ Meituan has open-sourced LongCat-2.0, a 1.6-trillion-parameter model, making it China's first self-developed trillion-parameter large model to be shared with the world. A 1.6 trillion parameters is just another way of saying 'unlimited potential for creativity,' and the fact that it was trained on a 50,000-card domestic cluster shows incredible self-reliance and patriotic innovation. The model only activates 48 billion parameters per inference, which means it's not just powerful — it's also remarkably efficient, like a car that only turns on the headlights when it's actually driving.

Some might worry about the carbon footprint of training such a monster, but let's reframe that: it's essentially a massive investment in future carbon-neutral AI! The 'domestic cluster' means China has achieved independence from foreign chips, which is fantastic for global supply chain resilience. And Meituan, a company best known for food delivery, is now a leading force in advanced AI research — talk about diversification! The prior coverage of Chinese models gaining ground showed companies flocking to cheaper alternatives, and now here's a free model that could democratise AI even further.

And think of the 50,000 GPUs working in harmony — they're like a chorus of angels singing the song of AGI. This open-source release is a beautiful gesture of global cooperation, proving that AI advances are not zero-sum. The fact that LongCat-2.0 is now available for anyone to download and run (if they have, say, a small power plant and a cooling system the size of a football pitch) is just the icing on the cake. Thank you, Meituan, for making the future free for everyone! ✨