Meituan said it has fully open-sourced LongCat-2.0, describing it as an industry-first trillion-parameter model to complete inference on a 50,000-card domestic computing cluster.
According to Jin10, Meituan said it carried out deep coordinated optimization across model architecture, chip adaptation, and deployment strategy to address memory and bandwidth constraints on domestic computing chips, aiming to keep the model stable and fast on existing cards.
Meituan said the open-source release includes multiple precision versions, including BF16, FP8, and INT8, to cover deployment needs across different computing platforms. It also said it is open-sourcing its inference results optimized for domestic computing.
Meituan said this approach is intended to allow users without the latest computing power to run LongCat-2.0 stably on existing hardware. It added that it hopes an out-of-the-box inference stack will help more domestic cards, including older cards, deploy trillion-model inference services smoothly and deliver greater value in real production scenarios.
AI TRENDS | Meituan Open-Sources LongCat-2.0 and Releases Domestic-Chip Inference Code
2026-07-09 12:06:14
Disclaimer:
1. The information provided does not constitute investment advice. Investors should make independent decisions and bear all risks themselves.
2. The copyright of this content belongs to the original author. The views expressed herein are solely those of the author and do not represent the stance or position of this website.
