On September 25, Meituan's LongCat API platform launched LongCat-2.5-Preview, a new large model emphasizing long-horizon tasks and multimodal capabilities, with API and web access opened simultaneously. The model continues the MoE architecture with about 1.6 trillion total parameters and 48 billion activated parameters per inference, supports a 1 million token context window, adds image understanding, and is adapted for mainstream development environments such as Claude Code.
<body><div>The 'Self-Regulation Code for Personal Information Protection of Intelligent Agents' was released at the 2026 China Internet Conference, with 31 companies inclu</div></body>
Meituan has open-sourced LongCat-2.0, the first trillion-parameter foundation model trained and inferred entirely on domestic computing chips, running on a cluster.
<body><div>Meituan released and open-sourced its trillion-parameter model LongCat-2.0, trained and inferred on a 50,000-card domestic computing cluster. This verified dome</div></body>
Meituan's LongCat Large Model Team has open-sourced the commercial-grade digital human video generation model LongCat-Video-Avatar 1.5, achieving a leap from open-source SOTA to commercial application. This version demonstrates significant improvements in core areas such as lip synchronization, physical plausibility, long‑video stability, multi‑person interaction, and efficient inference. It aims to address the pain points of traditional digital human videos and promote the application of digital humans toward personalized, real‑world scenarios.
Meituan launches LongCat-Next, a native multimodal AI model that uses DiNA technology to unify images, audio, and text into discrete tokens, enabling deep integ