Tech Brief (Oct. 28): Meituan Releases Open-Source Video Generation Model
Listen to the full version

Meituan releases open-source video AI model
Meituan on Oct. 27 released an open-source video generation model named LongCat-Video. The model has 13.6 billion parameters and is based on the Diffusion Transformer (DiT) architecture. Meituan said the model can stably generate videos up to five minutes long without quality loss, thanks to pre-training on video continuation tasks. It is designed for scenarios requiring long-term dynamic simulation, such as digital humans and embodied AI.
Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.
Subscribe to both Caixin Global and The Wall Street Journal — for the price of one.
- DIGEST HUB
- Meituan released LongCat-Video, a 13.6B-parameter open-source video AI model generating up to five-minute videos.
- WeRide announced a Hong Kong IPO, issuing 88.25M shares at up to HK$35 each, with trading starting Nov. 6.
- China’s 5G base stations reached 4.705M by September; telecom revenue in the first three quarters was 1.327T yuan ($186.6B), up 0.9% year-on-year.
- Meituan
- Meituan, a Chinese tech company, released an open-source video generation model named LongCat-Video on October 27. This model, boasting 13.6 billion parameters and built on the Diffusion Transformer (DiT) architecture, can generate videos up to five minutes long without quality degradation. LongCat-Video is designed for applications like digital humans and embodied AI.
- WeRide Inc.
- WeRide Inc., an autonomous driving firm, has revealed its plan for a Hong Kong public offering. The company intends to issue 88.25 million H-shares globally, priced up to HK$35 ($4.50) per share. Pricing is anticipated on November 4, with trading commencing on November 6.
- XPeng Inc.
- XPeng Inc. announced its global expansion in October, entering the markets of Lithuania, Latvia, Estonia, and Cambodia. In the Baltics, XPeng is introducing models like the G6 and G9, collaborating with local dealers. In Cambodia, six models were launched, with deliveries starting in November. XPeng's sales and service network now spans over 49 countries and regions.
- MiniMax
- MiniMax (月之暗面), a Chinese AI unicorn, launched M2, a new open-source large inference model. M2 uses a Mixture-of-Experts (MoE) architecture with 230 billion parameters (10 billion active) and an output speed of 100 tokens per second. It is designed for developing AI agents.
- PODCAST
- MOST POPULAR





