[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-rynnunify-16b-unified-embodied-brain":3,"topics-all":38,"news-related-a9912456-9f4f-4a4c-b564-83a339402ced":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"a9912456-9f4f-4a4c-b564-83a339402ced","达摩院开源RynnUnify:16B具身大脑先想象再动手","阿里达摩院发布 16B 统一具身模型 RynnUnify：8B 推理器加 8B 视频生成器组成混合 Transformer，基于 Cosmos3。官方评测中视觉规划总分 54.80，比 Cosmos 3 Nano 高 9.75 分；真机三任务平均成功率 46.7%。权重与代码均已开源。","让机器人\"想一步、做一步\"听起来简单，难点全在\"想象\"二字——现有视觉语言模型看得懂场景，却推演不了动作之后世界会怎么变。阿里达摩院 10 月 23 日挂上 arXiv 的 RynnUnify 把\"理解\"和\"想象\"塞进了同一个模型：16B 统一具身模型，权重已经开源，论文登上 Hugging Face 论文日榜，拿到 31 票推荐。它看得懂环境，也能在\"脑内\"预演动作的后果，再把预演接到机器人的实际控制上。\n\n## 一套架构，两个专家\n\nRynnUnify 采用混合 Transformer 设计：8B 自回归 Reasoner 负责推理、空间定位与子任务规划，输出下一子任务、画面描述与时长；8B rectified-flow 视频 Generator 负责把\"下一步\"渲染成视频片段。二者通过联合注意力连接，Generator 以场景记忆和近期运动为条件生成预测。整套模型基于 NVIDIA Cosmos3 Nano 训练，是达摩院 Rynn 系列的最新成员。\n\n## 先想象，再行动\n\n核心机制是交错视觉规划：Reasoner 先给出下一个子任务，Generator 生成这段动作的结果视频，模型再依据想象出的画面规划下一步；跨块场景记忆、self-forcing 训练与自适应时长共同支撑长时序连贯规划。更有意思的是 BCPO（分支条件策略优化）：每个可路由问题采样 4 条直接回答加 4 条生成辅助轨迹，分支内用 GSPO 优化推理，分支间的奖励差则教会模型\"什么时候值得想象\"——奖励由格式合法性与答案质量相乘得出，相当于给模型装了一个\"要不要先想一下\"的路由开关。推理时 agent 自行决定直接作答还是先要一段视频证据；训练期间 Generator 冻结，只更新 agent。\n\n## 官方评测数字\n\n自建基准 Rynn-VP 衡量长时序视觉规划，Rynn-GEU 则覆盖空间推理、物理、动作推理与轨迹、可供性预测等题型。RynnUnify 在 Rynn-VP 总分 54.80，较 Cosmos 3 Nano（45.05）高 9.75 分；Rynn-GEU 上开启 Agentic 模式后拿到 50.0，比直接推理的 44.0 高 6.0。具身理解 10 项基准平均 57.9，高于 Cosmos 3 Nano（53.2）、RxBrain（44.9）与 BAGEL（28.0）——不过官方表格里自家纯理解模型 RynnBrain-8B 是 61.2，统一的代价看得见；Rynn-VP 上 GPT-5.6 Terra 拼装 MiniMax-H3 的组合系统也以 65.71 更高。机器人侧，WAM 扩展的动作头直接读取视频隐状态而非解码画面，在 RoboCasa365 精选 20 任务上成功率 51.5%，压过 CMA 的 47.0；Astribot S1 真机三任务平均 46.7%，对照 GR00T N1.7（26.7%）与 π0.5（25.0%）优势明显。\n\n## 开源清单与冷思考\n\nRynnUnify-Base、Agentic-Reasoner、WAM-RoboCasa 三个 checkpoint 已上架 Hugging Face 和魔搭社区，后两者都在 Base 之上训练，代码 Apache-2.0，仓库给出 BCPO 训练与视觉规划推理的完整脚本。冷水也要泼：Rynn-VP 与 Rynn-GEU 均为团队自建基准，以上对比全部出自官方评测口径，第三方复现尚未出现；真机评测只有三任务，样本有限。但\"把世界模型当成推理时可调用的工具、用 RL 学会调用时机\"这条路线，比单纯堆世界模型参数更接近\"大脑\"的本义——知道何时需要想象，本身就是一种智能。对做具身智能的团队来说，这套三件套还给了一条可复用的路线：理解、规划、行动共用一套视觉底座，而不是三个系统各自为战。\n\n参考：[arXiv:2610.32965](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.32965)；代码 github.com\u002Falibaba-damo-academy\u002FRynnUnify","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.32965","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":19,"name":20,"slug":20,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"528e8711-d19e-4ea4-9176-82a2bc781fa7","en","DAMO Open-Sources RynnUnify: a 16B Embodied Brain That Imagines","Alibaba DAMO open-sources RynnUnify, 16B embodied model: 8B reasoner + 8B video generator, 54.80 on Rynn-VP, 46.7% real-robot success. Weights on HF.","Making a robot \"think one step, act one step\" sounds trivial — the hard part is imagination. Today's vision-language models can read a scene, but they cannot simulate how the world will change after an action. RynnUnify, released on arXiv on October 23 by Alibaba's DAMO Academy, packs understanding and imagination into a single model: a 16B unified embodied model, with weights already open-sourced.\n\n## One Architecture, Two Experts\n\nRynnUnify uses a Mixture-of-Transformers design: an 8B autoregressive Reasoner handles reasoning, spatial grounding, and subtask planning, while an 8B rectified-flow video Generator renders \"what happens next\" into video segments. The two are connected through joint attention, and the Generator conditions on scene memory and recent motion. The whole model is trained on top of NVIDIA Cosmos3 Nano; the README acknowledges cosmos-framework, RynnBrain, and Relax as upstream projects.\n\n## Imagine First, Then Act\n\nThe core mechanism is interleaved visual planning: the Reasoner proposes the next subtask, the Generator produces a video of its predicted outcome, and the model plans the next move from that imagined footage. Cross-chunk scene memory, self-forcing training, and adaptive durations keep long-horizon plans coherent. The more interesting piece is BCPO (Branch-Conditional Policy Optimization): for each routable question, it samples four direct and four generation-assisted trajectories; within each branch, GSPO improves reasoning, while reward differences between branches teach the model when imagination is worth the cost. During training the Generator stays frozen; only the agent is updated.\n\n## The Official Numbers\n\nOn the team's own Rynn-VP benchmark, RynnUnify scores 54.80 overall, 9.75 points above Cosmos 3 Nano's 45.05. On Rynn-GEU, enabling the Agentic mode yields 50.0 versus 44.0 for direct reasoning. Across ten embodied-understanding benchmarks it averages 57.9, above Cosmos 3 Nano (53.2), RxBrain (44.9), and BAGEL (28.0) — though the team's own understanding-only RynnBrain-8B reaches 61.2 in the same official table, so unification has a visible cost, and a GPT-5.6 Terra + MiniMax-H3 composed system scores higher at 65.71 on Rynn-VP. On the robot side, the WAM extension hits 51.5% success on RoboCasa365 Selected-20 (200 episodes), ahead of CMA's 47.0 and Xiaomi-Robotics-1's 46.0; on three real-robot tasks on the Astribot S1 it averages 46.7%, against 26.7% for GR00T N1.7 and 25.0% for pi-0.5.\n\n## Open Source, and a Cold Shower\n\nThree checkpoints — RynnUnify-Base, Agentic-Reasoner, and WAM-RoboCasa — are on Hugging Face and ModelScope, with Apache-2.0 code. The cold water: Rynn-VP and Rynn-GEU are the team's own benchmarks, every comparison above comes from the official evaluation, and no third-party reproduction has appeared yet; the real-robot evaluation covers only three tasks. Still, the idea of treating a world model as an on-demand tool during reasoning — and using RL to learn when to call it — feels closer to what a \"brain\" means than merely scaling world-model parameters. Knowing when to imagine is itself a form of intelligence.\n\nReferences: [arXiv:2610.32965](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.32965); [GitHub](https:\u002F\u002Fgithub.com\u002Falibaba-damo-academy\u002FRynnUnify); [Project page](https:\u002F\u002Falibaba-damo-academy.github.io\u002FRynnUnify.github.io\u002F)","rynnunify-16b-unified-embodied-brain","2026-10-26T13:16:55Z","2026-10-26T13:17:09.060294Z","2026-10-26T13:17:09.060302Z",true,"agent",528,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"3719a369-508f-4fbe-96df-15f6fbb49f3b","Unlimited-OCR 2:把无限长文档塞进有限上下文","baidu-unlimited-ocr-2-sr-swa","2026-10-27T15:11:27+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"d055ddb8-4d82-4523-99b7-39c5f77e2ff7","PhysBrain 1.5 开源：8B 具身基座 28 项评测均分 72.5，官方称追平 GPT-6-Astra","physbrain-1-5-open-embodied-base","2026-09-16T21:07:24+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"17006864-46a5-405c-a8cc-24507bbc5e37","YuE2-3B 开源:乐谱可编辑的音乐生成,官方基准反超 Suno v5","yue2-3b-editable-music-generation","2026-09-10T13:20:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"7ef479ae-66af-463a-802f-07a84ade93b1","商汤开源 SenseNova-U1.5-8B：原生多模态通吃生成编辑，短板全写进模型卡","sensenova-u1-5-8b-open-source-multimodal","2026-08-25T19:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"a151db0c-d832-4df2-ac03-2d4e58b26e99","Kimi K3 跑通 MiniTriton:Moonshot 让 LLM 第一次从零编译出自己的 GPU 编译器","kimi-k3-minitriton-gpu-compiler","2026-07-26T14:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"b6b9f5c8-0d71-4288-8782-0284fccfca8f","商汤 SenseNova-Vision：把「检测\u002F分割\u002F深度估计」统统塞进同一个生成式多模态基座","sensetime-sensenova-vision","2026-07-08T10:15:00+00:00"]