Meituan LongCat Releases LongCat-2.5 Multimodal Model

It enhances native visual understanding and agent programming via extended trajectory data and large-scale reinforcement learning. Visual inputs are encoded as hierarchical discrete tokens, enabling unified processing with text. Real agent failure trajectories are converted into reproducible training tasks to address performance gaps.

Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.