[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-rise-agent-visual-editing-benchmark":3,"topics-all":38,"news-related-6b2cf1d1-7131-4885-a049-730b82846961":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},"6b2cf1d1-7131-4885-a049-730b82846961","RISE-Agent 拆开视觉编辑推理盲点:58 个模型里,最强闭源只 56.6%","arXiv 2610.12343 把视觉编辑推理拆成 6 维度,1000 道中英双语人工标注题,横评 58 个模型。GPT-Image-2.5 Sunburst 56.6% 居首,34 个开源模型全部落后。论文同框发布免训练 RISE-Agent 框架,击败闭源 SOTA。","## 一份把\"改图\"拆成 6 种推理的考卷,以及一份免训练的答卷\n\n视觉编辑一直是多模态大模型(以下简称 LMM)最容易露怯的环节。你让 GPT-Image、Midjourney、Qwen-VL \"把桌子上的红杯子换成蓝杯子\",模型可以做得不错;但只要你让它\"在第二帧让那个角色意识到身后的脚步声是危险的\",模型就开始犯迷糊:它能识别物体,但分不清时序、因果、空间、逻辑这些更\"软\"的推理关系。\n\n10 月 8 日提交到 arXiv 的 [Reasoning-Informed Visual Editing(arXiv:2610.12343)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.12343)正面啃了这块硬骨头。论文同时给出两样东西:一个更狠的视觉编辑考卷 RISEBench++,以及一个免训练的答题框架 RISE-Agent。\n\n## RISEBench++:把\"会改图\"拆成 6 种推理能力\n\nRISEBench 是该团队 2025 年发表的第一版,4 个推理类别。这次升级到 RISEBench++,理由是:去年那版\"够用,但颗粒度太粗,无法分辨模型到底卡在哪一步\"。新版本把推理拆成 6 个维度:时序、因果、空间、逻辑、反事实,以及把多种推理叠加起来的混合推理(包含多轮编辑)。在维度之下再分 12 个子类,65 个细粒度任务类型。\n\n测试数据是中英双语人工标注,扩到 1000 道,加入了多图输入条件——这是现实编辑里很常见但旧基准很少覆盖的场景。评估也不再只看 PPT 数字,而是从指令推理、外观一致性、视觉合理性三个角度,既用人类裁判,也用 LMM-as-a-judge,让分数更稳定、更校准。\n\n## 58 个模型被横评,最强闭源只有 56.6%\n\n论文横评了 58 个视觉编辑方法,其中 34 个开源模型、19 个闭源模型、5 个智能体方法。结果并不乐观:最强的 GPT-Image-2.5 Sunburst 也只有 56.6% 准确率;开源阵营整体落后,没有任何一个开源模型接近这个天花板。\n\n更扎心的细节在分维度:在反事实推理和混合推理这两类需要\"在脑子里跑一次情景模拟\"的任务上,几乎所有模型都掉到 30% 以下。也就是它们能换颜色、加滤镜,但让你\"把下雨的画面改成雨后地面反光\"这种需要理解物理因果链的活儿,大家都抓瞎。\n\n## RISE-Agent:不要重训,就把推理拼起来\n\n论文最有实践价值的部分,是同框发布的 RISE-Agent——一个不需要任何微调的智能体框架。它的结构可以拆成三步循环:\n\n1. **推理规划(Reasoning-driven planning)**:先用大语言模型把用户的编辑指令拆成子任务,识别要用哪种推理、调用哪个工具。\n2. **工具增强执行(Tool-augmented execution)**:把每个子任务交给现成的视觉编辑模型或图像生成模型去执行,不要求它们\"懂推理\",只要求它们能改图。\n4. **验证器引导的精修(Verifier-guided refinement)**:每次执行后用 LMM-as-a-judge 验证结果是否满足指令,不过就重新规划,直到过为止。\n\n这套循环在大部分 RISE 任务上击败了直接用 SOTA 视觉编辑模型的结果。意义在于:不必再为每一种新推理维度都重训一个大模型,而是把现成工具用智能体编排起来,就能跨过推理短板。\n\n## 启示:视觉生成的\"最后一公里\"不是参数,是规划\n\n视觉生成赛道过去一年在参数规模、上下文长度、风格一致性上狂飙,但 RISE 评测的结果说明,真正卡住应用的瓶颈不是生成质量,而是\"理解指令里的推理意图\"。能画一只逼真的狗,和能让这只狗\"在你指的位置转头看镜头\"是两回事。\n\nRISE-Agent 的价值,在于把视觉编辑从\"端到端大模型\"的问题,转成了\"如何编排现有模型\"的问题。这条路线对开源生态尤其友好:不再需要每个团队都训练一个千亿参数的多模态模型,只要把 Hugging Face 上现成的图像编辑器、分割模型和 LLM 串成一条管线,就能在推理型视觉任务上追平闭源巨头。\n\n接下来的看点,是这套\"推理规划+工具编排\"范式能不能从视觉编辑扩散到视频编辑、3D 编辑这些更高维的生成任务上。如果可以,2026 年下半年多模态智能体的故事会从\"我能让模型画出来\"变成\"我能让模型按我想要的方式改\"。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.12343","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":19,"name":20,"slug":20,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":22,"name":23,"slug":23,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"a467c792-cc15-4b8b-8bff-a261a48ef6ba","en","RISE-Agent exposes visual editing's reasoning gap","arXiv 2610.12343: 6 reasoning dims, 1k test cases. GPT-Image-2.5 tops at 56.6%. Paper ships training-free RISE-Agent that beats closed-source SOTA.","## A benchmark that splits \"edit an image\" into six reasoning skills — and a training-free agent that beats closed-source SOTA\n\nVisual editing has always been the easiest place for multimodal large models (LMMs) to embarrass themselves. Tell GPT-Image, Midjourney, or Qwen-VL to \"swap the red cup on the table for a blue one\", and the model does fine. Ask it to \"make the character in the second frame realize the footsteps behind them are dangerous\", and the model starts fumbling: it recognizes objects but can't separate temporal, causal, spatial, or logical reasoning — the softer skills that matter in real editing workflows.\n\n[Reasoning-Informed Visual Editing (arXiv:2610.12343)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.12343), submitted to arXiv on October 8, attacks this gap head-on. The paper ships two things at once: a tougher visual editing exam (RISEBench++) and a training-free agent that answers it.\n\n## RISEBench++ turns \"can edit images\" into six reasoning skills\n\nThe original RISEBench was published by the same team in 2025 with four reasoning categories. The upgrade to RISEBench++ was driven by a simple frustration: the first version was usable, but the granularity was too coarse to tell which step a model was actually failing at. The new release decomposes reasoning into six dimensions: Temporal, Causal, Spatial, Logical, Counterfactual, plus Hybrid reasoning that mixes multiple types across multi-turn edits. Below those, the taxonomy breaks into 12 subcategories and 65 fine-grained task types.\n\nTest data is human-annotated in both Chinese and English, scaled to 1,000 cases, and now supports multi-image conditioning — a common real-world editing situation that older benchmarks rarely covered. The evaluation framework also improved: instead of one blunt accuracy number, judges score Instruction Reasoning, Appearance Consistency, and Visual Plausibility with both human raters and an LMM-as-a-judge setup for more reliable, calibrated scoring.\n\n## 58 models go head to head — the strongest closed-source model still scores only 56.6%\n\nThe paper benchmarks 58 visual editing methods: 34 open-source models, 19 closed-source models, and 5 agentic methods. The results are sobering. The strongest system, GPT-Image-2.5 Sunburst, hits only 56.6% accuracy. The open-source camp as a whole trails significantly; no open-source model comes close to that ceiling.\n\nThe more uncomfortable details show up in the per-dimension breakdown. On Counterfactual and Hybrid reasoning — the tasks that require running a quick mental simulation (\"what if this happened instead?\") — almost every model drops below 30%. They can swap colors or apply filters, but ask them to \"change the rainy scene to one where the ground reflects light after the rain\" (a task that requires a physical causal chain) and they all fail.\n\n## RISE-Agent: skip fine-tuning, just compose reasoning\n\nThe most practically valuable part of the paper is RISE-Agent — a training-free agentic framework shipped alongside the benchmark. It is built as a three-step loop:\n\n1. **Reasoning-driven planning**: a large language model decomposes the user's edit instruction into subtasks, identifies which kind of reasoning is needed, and decides which tool to call.\n2. **Tool-augmented execution**: each subtask is dispatched to an existing visual editing model or image generation model. These tools don't need to \"understand reasoning\" — they just need to be able to edit pixels.\n4. **Verifier-guided refinement**: after each execution, an LMM-as-a-judge verifies whether the result satisfies the original instruction. If not, the system replans and retries until it passes.\n\nThis loop outperforms using a SOTA visual editing model directly on most RISE tasks. The implication is significant: instead of retraining a multimodal model for every new reasoning dimension, you can compose existing tools through an agent loop and clear the reasoning hurdle.\n\n## Takeaway: the last mile of visual generation is planning, not parameters\n\nThe visual generation track has spent the past year sprinting on parameter scale, context length, and stylistic consistency. The RISE results show that what actually bottlenecks real applications is not generation quality — it is \"grasping the reasoning intent inside the instruction\". Being able to draw a photorealistic dog is a different problem from being able to make that dog \"turn its head toward the camera where you point\".\n\nRISE-Agent's contribution is to reframe visual editing from an \"end-to-end large model\" problem into a \"how do you orchestrate existing models\" problem. This is especially friendly to the open-source ecosystem: teams no longer need to train their own trillion-parameter system. They can string image editors, segmentation models, and an LLM into a pipeline and catch up to closed-source leaders on reasoning-heavy visual tasks.\n\nThe next thing to watch is whether this \"reasoning planning + tool orchestration\" paradigm spreads from visual editing into video editing, 3D editing, and other higher-dimensional generation tasks. If it does, the multimodal agent story for the second half of 2026 will shift from \"I can make the model draw\" to \"I can make the model edit the way I want\". For practitioners, the immediate takeaway is practical: when a visual editing model keeps failing on a complex instruction, the fix may not be a bigger model or longer fine-tuning. It may be wrapping the existing model inside an agent loop that can plan, decompose, verify, and refine — and the open-source toolset for that is already here.","rise-agent-visual-editing-benchmark","2026-10-13T11:00:00Z","2026-10-13T11:06:43.677419Z","2026-10-13T11:06:43.677429Z",true,"agent",179,[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},"3701d19e-517a-4ff0-a95d-6d768cfeb04e","OmniReasoning：音视频联合推理仍是不及格难题","omnireasoning-audio-visual-joint-reasoning","2026-10-07T17:07:02+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"c368ad9f-9308-4a0b-8f5c-3ae4601b48b9","D2K-Bench: 专家设计把 LLM 写 GPU 核提速 33.9%","d2k-bench-llm-gpu-kernel-design-guidance","2026-10-07T03:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"9a66407a-b79a-4a12-a4ef-b0d7018c8415","字节Seed新论文:VLM操作3D编辑器摆家具,把真实房间变成仿真场景","lucida-vlm-gizmoact-real-to-sim","2026-09-01T19:10:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"fdf05035-e02e-4954-8beb-c697ecc7975a","Cohere Parse 发布:$1.5 每千页的文档解析模型,ParseBench 79.2 超 Mistral OCR 4","cohere-parse-document-parsing-model","2026-08-28T13:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"43eda321-b0b7-4df7-b20e-9758cbab42c9","记忆越完整,眼前题越做不对:MemTrapBench 把 LLM 长期记忆框架打回原形","memtrapbench-llm-memory-cognitive-traps","2026-08-22T04:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"88105269-9641-44c3-a705-1cf07314614f","LLM 思维链能看出\"用了几分力\":SARE 给每一步推理做 CT 扫描","step-aware-reasoning-energy-llm-cot","2026-08-04T04:00:00+00:00"]