AIsaAIsaStudio English 打开 AIsa Studio

美团 LongCat 发布 LongCat-2.5:1.6 万亿参数原生多模态 MoE 模型

原标题:meituan-longcat/LongCat-2.5 (image-text-to-text)

美团 LongCat 团队在 Hugging Face 发布 LongCat-2.5,这是一款面向长程智能体任务的原生多模态 MoE 模型,总参数量 1.6 万亿,每 token 激活约 48B,采用 MIT 许可。模型在 LongCat-2.0 基础上结合视觉理解与智能体编码,训练采用长序列 SFT、大规模 RL 和多教师 KL 蒸馏,并报告长序列 SFT 任务中单卡吞吐较 DDP 基线提升 86%。评测显示 Terminal Bench v2.1 得分 88.4(LongCat-2.0 为 70.8)等多项对比成绩。

为什么值得看

README 给出参数规模、训练方法和与多款前沿模型的评测对比,读者可据此评估它在智能体编码任务上的实际位置。

阅读原文 ↗

本页的摘要、推荐理由与解读出自 AIsa Studio 的 AI 风向智能体,原文版权归原作者所有。

原文

Hugging Face 热门模型第 29 名:meituan-longcat/LongCat-2.5,241 赞,近期下载 971 次。 标签:safetensors, pro_v, multimodal, image-text-to-text, text-generation, mixture-of-experts, agentic, conversational, custom_code, arxiv:2603.27538, arxiv:1909.07606, license:mit

README:

LongCat-2.5

Tech Blog 📄

Model Introduction

LongCat-2.5 is a native multimodal MoE model for long-horizon agentic tasks, with 1.6 trillion total parameters and approximately 48 billion activated per token. Building on LongCat-2.0, it combines visual understanding with agentic coding for complex codebase development, full-project builds, creative design, and 3D tasks. Long-sequence training, large-scale RL, multi-teacher learning, and infrastructure optimizations on domestic AI accelerators underpin these capabilities.

LongCat-2.5 Highlights

  • Native Visual Modeling: Building on LongCat-Next, native-resolution visual features are converted into discrete tokens and jointly aligned with text in a shared semantic space.

  • Long-Horizon Agentic RL: Training combines real-world engineering tasks and reproducible agent failures with efficient rollout strategies and safeguards against reward hacking, improving reliable end-to-end task execution.

  • Multi-Teacher Learning: Specialized agent, reasoning, and interaction teachers guide the model through full-vocabulary KL distillation and Tree Attention, strengthening multi-turn execution, reasoning, instruction following, and safety.

  • Efficient Long-Context Training: Supernode-aware parallelism, efficient sharding, and communication–computation overlap improve training efficiency, achieving 86% higher per-device throughput than the DDP baseline in reported long-sequence SFT tasks.

  • RL Infrastructure: Drawing on community practices, aligned training and inference logic maintains bitwise-identical token log probabilities across training, prefill, and decoding. KV cache optimization and LSA improve serving efficiency, nearly doubling decode KV capacity without increasing TPOT on the tested workload.

For full technical details, see our blog.

Evaluation Results

We evaluate LongCat-2.5 across coding, general agent, and visual tasks. Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only. - indicates an unreported score. See the expandable footnotes below for evaluation settings.

BenchmarkLongCat-2.5LongCat-2.0GPT-5.5GPT-5.6-SolClaude-Opus-4.8Claude-Opus-5GLM-5.3kimi-k3DeepSeek-V4.1-Flash

Code Agent Terminal Bench v2.188.470.883.288.874.689.188.288.390.6* Terminal Bench v4.038.53.0-37.323.653.941.812.631.2 Program Bench (Almost Resolved)25.52.513.523.0<t

原文版权归 Hugging Face 热门模型 及原作者所有,本站转载仅供阅读。如需删除,请联系 support@aisa.one。

事件综述

美团开源 1.6 万亿参数多模态 MoE 模型 LongCat-2.5

2026年11月14日,美团 LongCat 团队在 Hugging Face 发布开源模型 LongCat-2.5,定位为面向长程智能体任务的原生多模态 MoE 模型:总参数量 1.6 万亿,每 token 激活约 48B,权重采用 MIT 许可。报道称模型在 LongCat-2.0 基础上结合视觉理解与智能体编码,训练采用长序列 SFT、大规模 RL 和多教师 KL 蒸馏,并报告长序列 SFT 任务中单卡吞吐较 DDP 基线提升 86%;评测方面 Terminal Bench v2.1 得 88.4 分(LongCat-2.0 为 70.8)。11月15日 The Decoder 跟进报道,补充称该模型完全使用中国产芯片训练,示例部署使用 194 张华为 Atlas A2 昇腾加速器;API 支持最多 100 万 token 输入和 12.8 万 token 输出,兼容 Claude Code、Codex 和 Hermes 等智能体框架,并设有限时促销价约每百万输入 token 0.27 欧元、输出 1.07 欧元。该报道还称模型在 BrowseComp 上得 91.9 分,Terminal Bench v2.1 得 88.4 分接近 Claude Opus 5 的 89.1 分。两篇报道均来自二级来源,上述成绩与训练细节为发布方或转述方说法,尚未经独立验证。