Meituan Open-Sources LongCat-Video, Advancing China's Generative AI Capabilities

Meituan, the Chinese technology giant, has open-sourced LongCat-Video, an advanced AI model capable of generating high-definition videos up to five minutes long from text or images. The release, announced by the @Meituan_LongCat team, marks a significant development for China’s generative video technology landscape. Pandaily highlighted the news, stating, "> @meituan open-sources LongCat-Video, an AI model that generates 5-minute HD videos from text or images — a leap for China’s generative video tech."

LongCat-Video is a foundational model with 13.6 billion parameters, designed to handle Text-to-Video, Image-to-Video, and Video-Continuation tasks. Built on a Diffusion Transformer (DiT) architecture, the model is natively pretrained for video continuation, enabling it to produce minutes-long videos without color drifting or quality degradation, a common challenge in AI video generation. It generates 720p, 30fps videos efficiently, utilizing a coarse-to-fine strategy and Block Sparse Attention.

The open-sourcing of LongCat-Video under an MIT license makes it freely available for both research and commercial applications, challenging proprietary models from major tech players globally. Meituan positions this release as a crucial step towards developing more advanced "world models" capable of learning real-world geometry, semantics, and motion. This move is expected to foster innovation within the open-source community and increase pressure on other developers to offer more accessible solutions.

This initiative follows Meituan's earlier foray into open-source AI with the LongCat-Flash-Chat large language model in September 2025. The company has been actively integrating AI into its internal tools and customer service, aiming to boost efficiency across its diverse services, which include food delivery, e-commerce, and local services. Meituan's strategic investment in AI underscores its ambition to evolve beyond its core delivery business into a frontier tech player in China's competitive AI race.

The release of LongCat-Video contributes to China's growing prominence in the generative AI space, with several companies actively developing and open-sourcing powerful AI models. This competitive environment is driving rapid advancements, pushing the boundaries of what AI can create in visual media. Meituan's commitment to open science with LongCat-Video is poised to accelerate further research and application development in the field of AI-driven video content creation.