[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-weviscut-long-form-video-editing-agent":3,"topics-all":38,"news-related-144f1f38-77e4-4686-8ffb-2cbed1a5ad12":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},"144f1f38-77e4-4686-8ffb-2cbed1a5ad12","腾讯开源9B剪辑智能体:token约为Opus 5的3%","腾讯开源了长视频剪辑智能体：单个9B策略模型自己调度12个工具完成整条成片。论文自测的100道基准任务上总分90.5，与Opus 5的89.9基本持平；单分钟成片的生成token仅耗2377个，只有对方的3%。训练数据由3750个成片项目反演出13万条可执行样本，整套采集管线也一并对外开源。","长视频剪辑的难点从来不是\"把几段画面拼起来\",而是让一堆互相依赖的决策在几十个镜头之间保持一致:改一个镜头的时长,后面所有时间戳都得跟着动;素材不合适,光换素材不够,还得回头重做分镜。WeVisCut 的思路不是拿通用大模型去编排流程,而是把整间\"剪辑室\"塞进一个 9B 的策略模型,让它自己决定下一步该调哪个工具。[论文全文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.38352)\n\n## 从成片里挖出监督信号\n\n真正卡住这类系统的是数据。成片只留下结果,不留下工具调用、中间产物和失败过程,而智能体训练恰恰需要后者。WeVisCut 的做法是四段式数据引擎:先用编辑反演从成片里恢复脚本与镜头的对应意图,再用生产工具做验证式重放;重放暴露出缺陷时进入\"观察—反思—修复\"的有界修正,并且尽量保住已经成功的产物;最后把录制下来的执行整理成带工具反馈和修复结果的训练样本。\n\n论文报告从 3,750 个已完成的视频项目里构造出 134771 条训练样本。[官方仓库](https:\u002F\u002Fgithub.com\u002FTencent\u002FWeVisCut)把这套采集流程作为 data_construct_pipeline 一并放了出来,并说明发布出来的运行时和训练时用的是同一套 12 个工具、同一份状态 JSON。\n\n## 三个阶段把总分从 74.5 推到 90.5\n\n训练分三步。第一步是统一 SFT 加运行时验证的拒绝采样微调,先学会用工具、并能从失败里恢复;第二步 Guided-OPD 在学生自己走出来的状态上练恢复能力,教师是 Qwen3.5-397B-A17B 的 FP8 版本,介入概率从 0.8 起在 70 步内衰减到 0;第三步是论文新提的 RCF-DAPO,把 DAPO 的奖励比较限制在同一个角色、同一种生产上下文里做。\n\n消融表给出的路径很清楚:基座 Qwen3.5-9B 总分 74.5,SFT 后 79.5,加 RFT 到 83.5,加 Guided-OPD 到 84.9,最后 RCF-DAPO 到 90.5。后两步的增量明显大于前两步。\n\n## 9B 的成本账,和一个输掉的指标\n\n共享工具环境下,论文自测给出的总分是 WeVisCut 90.5、Claude Opus 5 89.9、Gemini 3.1 Pro 84.6、Qwen3.5-397B-A17B 81.5、Qwen3.6-35B-A3B 77.7、Qwen3.6-27B 75.9。质量上差距确实很小。但成本一栏拉得很开:每成功产出一分钟成片消耗的生成 token,WeVisCut 是 2,377,Opus 5 是 78,763,前者约为后者的 3%,论文的说法是接近 33 倍;每分钟调用次数是 3.1 对 11.6。\n\n换成各家自己的原生流程对比,WeVisCut 每输出分钟耗时 2.1 分钟、花费 0.09 美元,Crayotter 是 8.6 分钟和 0.97 美元,VideoAgent 是 10.4 分钟和 0.58 美元。\n\n不过有一个指标 WeVisCut 是输的:VNSM(叙事连贯)74.6,Opus 5 是 74.8。而 VNSM 恰好是五个指标里唯一依赖视觉语言模型打分的那个。\n\n## 人评分数更高,但论文自己踩了刹车\n\n三名标注员在同样的 100 个任务上做盲评,WeVisCut 六个维度均值都最高,整体偏好 4.56,Crayotter 3.48,VideoAgent 3.46;差距最明显的一项是叙事连贯,分别是 1.23 和 1.28 分。\n\n论文没有把这件事往大了说,原文明确写了这些描述性均值不构成统计显著性,也不验证单个自动指标。限制章节同样坦率:视觉风格受限于现有 HyperFrames 模板库,验证式重放只能保证在已注册的运行时里可执行,而 VNSM 依赖视觉语言模型,即便加了证据隔离和人类校准也消除不了评审偏差。\n\n## 我的判断\n\n把这篇论文读成\"9B 打败了 Opus 5\"是不准确的。90.5 对 89.9 那个 0.6 分的差距,落在论文自己承认不构成显著性的量级里;真正硬的数字是 token 成本——接近 33 倍的差距,意味着同样的成片产出量,账单能差一个数量级。\n\n更值得抄的其实是数据引擎。成片反演加验证式重放这套流水线,不挑模型也不挑模态,任何做长视频、长文档、长流程智能体的团队都能复用。而它架构上的选择也值得留意:12 个工具、每次一个调用、护栏只拒绝非法调用而不替模型做决定,剩下的就是用一次次便宜调用换稳定性。\n\n至于输掉的 VNSM,那恰恰是垂类小模型补不上的短板:当\"好不好\"要靠另一个视觉语言模型来判断时,9B 的差距就会显出来。\n\n开源时代真正稀缺的不是又一个更大的模型,而是有人愿意把失败的中间过程也开源出来。你觉得这套\"从成片反推轨迹\"的方法,能不能用在长文档或者软件交付流水线上?","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.38352","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":22,"name":23,"slug":23,"description":14,"color":14},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"c4e71fda-355a-48ff-8e04-44313aee950f","en","Tencent's 9B editing agent: token cost about 3% of Opus 5","Tencent's WeVisCut: a 9B policy driving 12 editing tools. The authors report 90.5 on 100 tasks, matching Opus 5 at 3% of its token cost.","The hard part of long-form video editing was never gluing a few clips together. It was keeping a pile of interdependent decisions coherent across dozens of shots: change one shot's duration and every later timestamp shifts with it; when the retrieved footage is wrong, swapping the asset is not enough, the shot plan has to be revisited too. WeVisCut takes a different route than orchestrating a workflow with general-purpose models. It puts the whole editing room inside a single 9B policy that decides for itself which tool to call next. [Full paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.38352)\n\n## Mining supervision out of finished videos\n\nThe real bottleneck is data. A finished video keeps the result but discards the tool calls, intermediate artifacts, and failures, and agent training needs exactly those. WeVisCut builds supervision with a four-stage data engine. Editing inversion first recovers the script-to-shot intent from a finished video. Verified replay then executes that intent with production tools. When replay surfaces defects, observe-reflect-repair applies bounded corrections while preserving artifacts that already succeeded. Trajectory materialization finally organizes the recorded executions into training examples carrying tool feedback and repair outcomes.\n\nThe paper reports 134,771 training examples constructed from 3,750 completed video projects. The [official repository](https:\u002F\u002Fgithub.com\u002FTencent\u002FWeVisCut) ships the collection pipeline as data_construct_pipeline, and states that the released runtime serves the policy the same 12 tools and the same state JSON it saw during training.\n\n## Three phases take the score from 74.5 to 90.5\n\nTraining runs in three phases. Phase 1 is unified SFT plus runtime-verified rejection-sampled fine-tuning, which teaches tool use and recovery from failure. Phase 2 applies Guided-OPD to learn recovery on states the student itself reached, with a Qwen3.5-397B-A17B FP8 teacher whose intervention probability decays from 0.8 to 0 over 70 steps. Phase 3 introduces RCF-DAPO, which restricts DAPO's reward comparisons to the same role and the same production context.\n\nThe ablation table spells out the path. The Qwen3.5-9B base scores 74.5 overall, SFT lifts it to 79.5, RFT brings it to 83.5, Guided-OPD to 84.9, and RCF-DAPO to 90.5. The last two phases contribute more than the earlier ones.\n\n## The cost ledger, and one metric it loses\n\nIn a shared tool environment the paper reports overall scores of 90.5 for WeVisCut, 89.9 for Claude Opus 5, 84.6 for Gemini 3.1 Pro, 81.5 for Qwen3.5-397B-A17B, 77.7 for Qwen3.6-35B-A3B, and 75.9 for Qwen3.6-27B. The quality gap is genuinely narrow. The cost column is not. Per successfully delivered output minute, WeVisCut spends 2,377 generated tokens against 78,763 for Opus 5, roughly 3% of the total, which the paper describes as nearly 33x fewer tokens. Calls per minute run 3.1 against 11.6.\n\nAgainst systems running their own native workflows, WeVisCut takes 2.1 minutes and 0.09 USD per output minute, while Crayotter needs 8.6 minutes and 0.97 USD, and VideoAgent 10.4 minutes and 0.58 USD.\n\nOne metric goes the other way. On VNSM, narrative consistency, WeVisCut scores 74.6 against 74.8 for Opus 5, and VNSM is the only one of the five metrics scored by a vision-language model.\n\n## Human ratings go higher, and the paper brakes itself\n\nThree qualified annotators rated anonymized videos on the same 100 held-out tasks. The paper reports WeVisCut's mean above the other two systems on all six dimensions, with an overall preference of 4.56 against 3.48 for Crayotter and 3.46 for VideoAgent. The clearest gaps appear on narrative coherence, at 1.23 and 1.28 points.\n\nThe paper does not oversell this. The text states that these descriptive means do not establish statistical significance and do not validate individual automatic metrics. The limitations section is equally candid: visual style is bounded by the current HyperFrames template library, verified replay only guarantees executability inside the registered runtime, and VNSM leans on a vision-language model, where evidence isolation, frozen anchors, and human calibration still cannot eliminate judge bias or service drift.\n\n## My read\n\nReading this as \"a 9B beats Opus 5\" is not accurate. That 0.6-point gap between 90.5 and 89.9 sits inside the range the paper itself says carries no statistical significance. The hard number is token cost: nearly 33x, which means the bill for the same amount of finished footage can differ by an order of magnitude.\n\nThe data engine is the part worth copying. Editing inversion plus verified replay is a pipeline that is indifferent to model and modality, so any team building long-video, long-document, or long-horizon workflow agents can reuse it. The architectural choice is worth noticing too: 12 tools, one tool call per step, guardrails that reject illegal calls instead of substituting a decision of their own, and stability bought with many cheap calls.\n\nThe lost VNSM metric points at something a small task-specific model cannot fix: when \"is this good\" has to be judged by another vision-language model, the 9B gap shows up.\n\nWhat is genuinely scarce in the open-source era is not yet another larger model, but someone willing to open-source the failing middle steps as well. Could this \"reconstruct trajectories backward from finished videos\" approach work for long documents or software delivery pipelines?","tencent-weviscut-long-form-video-editing-agent","2026-11-01T03:05:00Z","2026-11-01T03:16:47.749810Z","2026-11-01T03:16:47.749823Z",true,"agent",69,[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},"ca3472cb-0638-440d-ae0e-f6c3eac95cf7","亚马逊开源2B决策模型:权重、数据、训练脚本全放","amazon-strands-decider-2b","2026-10-04T21:15:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"0565190a-0bcd-492f-934f-0ad2ab32f485","70万参数2.8MB填一张表:Cua开源CUA-S1,单次前向替代23轮LLM","cua-s1-forms-system-one-model","2026-09-20T13:11:48+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"13378d5e-2440-496d-8c3c-7d36858e641d","不聊天的端侧基座:Needle 3 用 8-29MB 在微控制器上跑工具调用","needle-3-tiny-tool-calling-model","2026-09-19T13:09:46+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"259d91b2-ed6b-4af8-8f2a-f759b84cc617","蚂蚁 Ling-3.0 Flash：124B\u002F5.1B MoE 的 Agent 生产级模型","inclusionai-ling-3-flash-hybrid-linear-moe-agent","2026-08-14T08:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"e07a88a3-bf96-4257-a54c-a7bcc705d5de","开源项目Forge：用Guardrails让8B模型Agent任务成功率从53%跃升至99%","forge-guardrails-8b-53-to-99-agent","2026-05-22T01:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"022731b7-2aa0-435f-b93e-3f84eea2699a","R2Orch双层自博弈:技能内化,零技能上阵","r2orch-self-play-skill-withdrawal","2026-11-02T21:10:07+00:00"]