Meituan's LongCat team on November 2 released LongCat-2.5, a native multimodal model for long-horizon agent tasks with a total of 1.6 trillion parameters and about 48 billion activated per token, building on LongCat-2.0 with stronger native visual understanding and agentic coding.
The model adds more long-trajectory data in long-sequence training and applies large-scale long-horizon reinforcement learning to improve end-to-end engineering on complex codebases and full project construction. It represents visual signals as multi-level discrete tokens so the backbone processes vision and text in a unified token form, excelling at visual coding, creative design and 3D tasks.
Upgrades are backed by infrastructure optimized for long-context training and W4A4 inference on domestic computing power, with rollout optimization ensuring efficient large-scale long-horizon RL.
Source: DoNews / NetEase (Nov 2, 2026)
