美团正式发布了 LongCat-2.5，一个拥有 1.6 万亿参数的原生多模态模型。该模型专为长时序 Agentic 编码任务设计，每个 token 使用约 480 亿活跃参数。美团报告称，在 Terminal Bench v2.1 上取得了显著的性能提升，得分 88.4，高于之前的 70.8，但开放权重仍有待公布。 AI
影响 此次发布推进了复杂 Agentic 任务的原生多模态能力，有望提升 AI 处理长时序编码项目的能力。
排序理由 前沿实验室模型发布，附带系统卡。
在 Hacker News — AI stories ≥50 points 阅读 →
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
Meituan's LongCat-2.5 is a 1.6T native multimodal model with about 48B active per token, trained with long-horizon RL for agentic coding and optimized for W4A4 inference on domestic compute.
<p>Meituan has officially released LongCat-2.5, the latest generation of its native multimodal large model built for long-horizon agent tasks. The model uses a unified token representation for vision and text, activates roughly 48 billion parameters per token out of 1.6 trillion …
LongCat-2.5: Native Multimodality for Agentic Tasks https:// longcat.ai/blog/longcat-2.5/ # ai
Meituan has released LongCat-2.5, a natively multimodal model with 1.6 trillion parameters, about 48 billion active per token, trained with long-horizon reinforcement learning for agentic coding and optimised for 4-bit inference on domestic chips. Meituan reports 88.4 on Terminal…