88Z AI, 2 November — Meituan LongCat has released LongCat-2.5, a native multimodal model with 1.6 trillion total parameters and approximately 48 billion activated per token. Designed for long-horizon agent tasks, it enhances visual understanding and agent programming through long-sequence training and large-scale Agentic RL to improve complex code development and project building.
At a glance
- LongCat-2.5 features 1.6 trillion parameters, enhancing native vision and agent programming
- Improves complex code development via long-sequence training and Agentic RL
The update
Meituan LongCat has officially launched LongCat-2.5, a native multimodal model. With 1.6 trillion total parameters and approximately 48 billion activated per token, it is designed specifically for long-horizon agent tasks. LongCat-2.5 focuses on strengthening native visual understanding and agent programming performance to improve efficiency in handling complex tasks. Technically, the model enhances capabilities in complex codebase development and full project construction through long-sequence training, large-scale Agentic RL, and failure trajectory reproduction. For visual processing, it employs multi-level discrete tokens to unify image and text inputs, enabling more efficient integration and understanding of multimodal information.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 龙猫LongCatSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.