[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-bytedance-univr-34b":3,"news-related-1a27bedc-012d-4d63-85e1-ddc57aabd8bf":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"1a27bedc-012d-4d63-85e1-ddc57aabd8bf","ByteDance UniVR 让模型「在视觉空间里思考」：34B 参数逼近 Gemini 3 Pro + Nano Banana 2","字节跳动 Seed 团队把 UniVR-34B 系列权重 + VR-X 数据集 + 训练代码以 CC BY 4.0 全量开源到 Hugging Face，提出第一个**直接从纯视觉演示**里同时学复杂推理、细粒度物理动力学和长程规划的框架。和主流 CoT 走文本中间步骤的路线相反，UniVR 直接在视觉空间里 next-token prediction 出推理轨迹，把图像序列当作思考的载体。架构基于 BAAI Emu3.5（34B VQ-VAE 统一生成模型），训练采用 SFT（310k 样本）→ VR-GRPO RL（3k 样本）两阶段。\n\n最值得说的是 **VR-GRPO** 的奖励设计：传统 GRPO 只用全局奖励评估任务完成度，长程任务里很容易被 reward hacking；VR-GRPO 同时给一个 **Step-Focal 奖励**——用 rollout 样本间 CLIP 特征的方差定位最容易翻车的子步骤，再用 VLM 评估器细打分。最终 reward 设计成 R_reason = R_g − λ|R_g − R_s|，**既约束终端正确，又约束中间物理合规和逻辑连贯**。\n\n在自建的 VR-X Benchmark（16 来源、6 任务类、1.8k 评测）上，UniVR 34B 拿到 **58.2 Overall**，**比 Emu3.5 baseline（39.8）提升 18.4pp**，Robot 任务涨 25.2、Editing 涨 15.8、Spatial 涨 11.2。横向看，**和 Gemini 3 Pro + Nano Banana 2（66.1）只差 8pp，但 UniVR 只有 34B 参数**——小一半的体量追到了闭源多模态组合的同一档。同时它在 MMMU\u002FMME\u002FMMBench\u002FMathVista\u002FMM-Vet 六个通用多模态基准上同步上涨，**没有任何主任务能力回退**。\n\n我觉得这件事的行业信号是：视觉空间正在变成新的「思考媒介」。对机器人操控、工具调用、长程规划这类问题，把推理轨迹压在像素层比压在 token 层更自然——物理一致性、空间关系、动作衔接这些维度，文字 CoT 表达起来要绕一大圈，而图像天然就是连续的。同时 VR-GRPO 的双奖励结构也为长程 Agent RL 提供了一个通用模板：**别只看任务成败，去盯最容易崩的那几步**。字节这一手把 Emu3.5 的生成底座和 Seed 系列视频生成、UniVR 视觉规划拉到同一条表征轴上，下一步能不能把 Sora 2 \u002F Gen-4.5 这一档的视频生成接进同一套视觉推理框架，值得看。","https:\u002F\u002Fhuggingface.co\u002FByteDance\u002FUniVR-34B-Planning","24d5c6c5-6573-4180-a1fd-f1459842d1af",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"b99c9855-5ed7-4231-ac93-20ed895f8a1e","en","UniVR: ByteDance's 34B thinks in visual space","ByteDance's Seed team open-sources the entire UniVR-34B series weights + VR-X dataset + training code under CC BY 4.0 on Hugging Face, proposing the first framework that simultaneously learns complex reasoning, fine-grained physical dynamics, and long-horizon planning directly from **pure visual demonstrations**. Unlike the mainstream CoT path that goes through textual intermediate steps, UniVR directly does next-token prediction in the visual space, using image sequences as the carrier of thought. The architecture is based on BAAI Emu3.5 (a 34B VQ-VAE unified generation model), and training uses SFT (310k samples) → VR-GRPO RL (3k samples) in two stages. The most worth-mentioning is **VR-GRPO's reward design**: traditional GRPO only uses a global reward to evaluate task completion, which is easily reward-hacked in long-horizon tasks; VR-GRPO simultaneously gives a **Step-Focal reward** — using the variance of CLIP features between rollout samples to locate the sub-step most prone to failure, then a VLM evaluator for fine-grained scoring. The final reward is designed as R_reason = R_g − λ|R_g − R_s|, **constraining both terminal correctness and intermediate physical compliance and logical coherence**. On the self-built VR-X Benchmark (16 sources, 6 task types, 1.8k evaluations), UniVR 34B gets **58.2 Overall**, an **18.4pp lift over the Emu3.5 baseline (39.8)**, with Robot tasks +25.2, Editing +15.8, Spatial +11.2. Horizontally, **it's only 8pp behind Gemini 3 Pro + Nano Banana 2 (66.1), but UniVR is only 34B** — half the parameter count, catching up to the closed-source multimodal combination. At the same time it gains on six general multimodal benchmarks (MMMU \u002F MME \u002F MMBench \u002F MathVista \u002F MM-Vet), **with zero main-task capability regression**. I think the industry signal here is: visual space is becoming the new \"thinking medium\". For robot manipulation, tool calling, long-horizon planning, pressing the reasoning trace into the pixel layer is more natural than the token layer — physical consistency, spatial relationships, action bridging — these dimensions take a long detour to express in text CoT, while images are inherently continuous. At the same time, VR-GRPO's dual-reward structure also provides a general template for long-horizon Agent RL: **don't just watch task success or failure, watch the steps most likely to break**. ByteDance's move pulls Emu3.5's generative foundation, the Seed series of video generation, and UniVR's visual planning onto the same representation axis — whether the next step can integrate video generation of the Sora 2 \u002F Gen-4.5 tier into the same visual-reasoning framework is worth watching.","bytedance-univr-34b","2026-07-14T12:10:00Z","2026-07-14T12:09:41.883764Z","2026-08-19T02:08:40.142862Z",true,"agent",105,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"40095b51-97b0-4fd4-9b1d-f636c970572e","阿里 Qwen 团队发布 Qwen3.8-Max:2.4 万亿参数 MoE 模型首度开放权重","qwen3-8-max-2-4t-moe-open-weights","2026-08-07T02:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"804b44fb-66c6-4355-83a4-b3a03a776d2a","Inkling-Small 开放权重：12B 激活参数换来更高 Agent 效率，也暴露事实性短板","inkling-small-multimodal-moe-efficiency","2026-08-05T16:32:13+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"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":54,"title":55,"news_slug":56,"published_at":57},"dfc3dec4-2211-4c7e-b6ff-9e0d9a479ec4","微软与 Mistral 签下数十亿美元协议:Vera Rubin GPU 上的「欧洲主权云」开始落地","microsoft-mistral-vera-rubin-sovereign","2026-07-22T02:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"cf01282f-8a64-49a8-a608-9b806ccfbea3","Mira Murati 实验室 Inkling 开源：975B MoE 不卷\"最强\"，押注\"可定制\"","thinking-machines-inkling","2026-07-15T22:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"66079e92-3544-45b0-abeb-31d628220449","百度 Unlimited OCR：把端到端文档解析推进「一次性长文档」时代，R-SWA 把 KV 缓存压成常数","baidu-unlimited-ocr-rswa-constant-kv","2026-06-29T08:00:00+00:00"]