Article · AI translation
IT Home reported on November 2 that Meituan's LongCat today released LongCat-2.5, a native multimodal model for long-horizon agent tasks. The model has 1.6 trillion parameters in total and activates approximately 48 billion parameters per token, further strengthening native visual understanding and agentic programming capabilities compared with LongCat-2.0.
According to the introduction, LongCat has added more long-horizon trajectory data to long-sequence training and used large-scale long-horizon reinforcement learning to improve the model's end-to-end engineering capabilities in tasks such as development in complex codebases and building complete projects. The team continues to expand this training pipeline through data synthesis, environment construction, and mixed training across Harnesses. In addition, LongCat turns failure trajectories from real Agents into reproducible, verifiable training tasks and continues to add training data targeting the model's weaknesses.
Meanwhile, LongCat-2.5 represents visual signals as multilevel discrete tokens, allowing the backbone model to process visual and text inputs in a unified token format. The combination of native visual understanding and agentic programming capabilities enables the model to perform well in Visual Coding and support tasks such as creative design and 3D.
The company said these upgrades are supported by continuously evolving infrastructure: optimizations for long-context training and W4A4 inference on domestically produced computing hardware, along with training-inference numerical consistency and long-context rollout optimizations, support the efficient operation of large-scale long-horizon RL.
IT Home provides the evaluation results:
Try it online: https://longcat.chat
API access: https://longcat.chat/platform/product
GitHub, Hugging Face: Coming soon
Advertising statement: The external links in this article (including but not limited to hyperlinks, QR codes, and access codes) are intended to provide more information and save time in making selections. The results are for reference only. All IT Home articles containing external links include this statement.