[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-workflowarena-multi-agent-orchestration":3,"topics-all":38,"news-related-50fb4155-e244-4a2f-8ee7-5596db52da9f":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},"50fb4155-e244-4a2f-8ee7-5596db52da9f","多智能体编排大考:最强模型也只解决36.95%任务","WorkflowArena 用 145 道跨 8 领域的长程任务考多智能体的“编排”能力:派不派工、怎么分工,模型自己定。9 模型 25 配置同台,最强的 Opus 5.5 也只解决 36.95%。换蓝图:Opus 的蓝图让弱模型平均涨 6.41 分,更能打的 Astra,蓝图反而把两个模型带垮——会做题≠会指挥。","给一个 AI 智能体 4 小时预算和一份真实的专业工件,让它自己决定:这活儿一个人闷头干,还是拉起一队“分身”分头推进、最后拼成一份交付物?WorkflowArena 就是这样一间考场,而成绩最好的考生,完全交卷率也只有 36.95%。这份基准论文把过去只存在于架构图里的“多智能体编排”,变成了一场可以打分的考试。\n\n## 考的是组织能力,不是做题能力\n\nWorkflowArena 收录 145 道长程任务,横跨软件工程、企业运营、科研、形式数学、金融、法律、内容与设计、3D 制作 8 个领域、17 种任务形态,项目由 Navers Lab 与 Einsia.AI 资助。每道任务提供一个初始工作区,要求交付由相互依赖的部件组成的成品,再由隐藏的验证器打分——各部件单独看都正确、却没有整合成一致整体的答卷,过不了关。\n\n协议设计是它最特别的地方:模型拿到一个 Workflow 工具,最多同时拉起 16 个 worker,每个 worker 都是它自己的全新实例;要不要派工、派几个、给每个 worker 什么上下文,全部由模型现场拍板,组织工作消耗的时间也从同一个 4 小时预算里扣。纯顺序单干同样是合法工作流。于是每一份交付物,都能和产生它的组织方式对上号。\n\n## 成绩单:天花板 36.95%,偏科严重\n\n论文评了 9 个模型、25 种“模型+档位”配置:Claude Opus 5.5 与 Sonnet 5.5 跑在 Claude Code 里,GPT-6 Astra 与 GPT-6.1 Sol 跑在 Codex 里,另外还有 GPT-6 Luna、Kimi K3.1 Pro、GLM-5.3 Flash、DeepSeek V4.1 Flash 和 Doubao-Seed-Evolving。\n\n最强的配置是 Opus + Claude Code 最高档,平均完全解决率 36.95%,而且没有任何一种配置在全部 8 个领域同时领先。形式数学最接近饱和,最佳配置解决 95.83%;法律与金融的天花板分别只有 9.09% 和 18.75%。更刺眼的是“部分分”和“能交付”之间的落差:法律任务平均能拿 70.7 的连续分,完全过关率却只有 2.9%;Opus 高档在软件工程里连续分高达 96.98,完全过关率也只有 17.65%——模型写得出一篇大部分正确的分析,却很难交出一份全部达标的成品。评分上,内容与 3D 交付由 Gemini-3.8-Flash 按细则打分,其余领域用程序化验证器。中国模型在这张表上集体掉队:Kimi K3.1 Pro 16.48%、Doubao 10.84%、DeepSeek V4.1 Flash 6.08%、GLM-5.3 Flash 4.43%,与 Opus 差着明显身位。档位的敏感度也不一样:Opus 五档从 22.02% 拉到 36.95%,Astra 全程只在 27.34% 到 30.80% 之间波动。\n\n## 会做题不等于会指挥\n\n论文的关键实验是“换蓝图”:在 32 道高难度任务上固定执行模型(Kimi、DeepSeek、Luna、Doubao),只更换写工作流的模型。结果显示,Opus 写的工作流让四个执行模型全部提分,静态模式平均加 6.41 分——Kimi 从 35.75 涨到 46.61,DeepSeek 从 18.37 涨到 25.24。反观 Astra 的蓝图,静态模式平均掉 2.53 分,还同时拉低了 Kimi 和 DeepSeek,尽管 Astra 自己单打独斗(解决率 27.34%)远强于这四个执行者。把自己那摊活干好的能力,并不会自动转化成组织别人干活的能力。\n\n两种“管理风格”也被数据拍实:高努力档下,Opus 平均每任务只开 0.86 个工作流、每个工作流带 5.55 个 worker,属于“少批次、大团队”;Astra 平均开 2.09 个、每个只带 1.72 个 worker,约 99% 的工作流不超过 4 人,典型的“高频次、小分队”。\n\n## 所以呢\n\n这份基准把“编排能力”从营销话术变成了可测量、可迁移的资产:同一批弱模型,接上好蓝图就能白涨 6 分以上。成本账同样直白:Opus 高档 6.66 美元\u002F任务解决 34.11%,Astra 在 xhigh 档花 29.26 美元只解决 30.80%,成本贵 4.4 倍、成绩还低 3.31 个百分点。对做 Agent 的人来说,单模型跑分之外,组织能力是下一块战场;对选型的人来说,先想清楚你要雇的是“能干的员工”,还是“能带队的经理”。\n\n论文:https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.12856","https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.12856","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},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"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},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"e71be7a7-a34d-4d70-b484-f9a680ee3180","en","WorkflowArena: Best Agents Fully Resolve Just 36.95%","WorkflowArena tests orchestration on 145 tasks in 8 domains: best config resolves just 36.95%; Opus workflows lift weak executors 6.41 points.","Hand an AI agent a four-hour budget and a workspace full of real professional artifacts, then let it decide: grind through the job alone, or spin up a team of its own clones, farm out the work, and stitch the pieces into one deliverable? That is the exam room WorkflowArena builds, and the best student in the class fully completes only 36.95% of the tasks. The benchmark paper turns \"multi-agent orchestration\" — a phrase that usually lives in architecture diagrams — into something you can score.\n\n## It tests organization, not problem-solving\n\nWorkflowArena packs 145 long-horizon tasks into 8 professional domains — software engineering, enterprise operations, research, formal mathematics, finance, law, content and design, and 3D production — with 17 task forms; the project is funded by Navers Lab and Einsia.AI. Each task provides an initial workspace and demands a deliverable made of interdependent components, graded by a hidden verifier. A submission whose parts are individually correct but never reconciled into a consistent whole does not pass.\n\nThe protocol is the interesting part. The model gets a Workflow tool that launches background workflows with up to 16 concurrent workers, and every worker is a fresh instance of the agent itself. Whether to delegate at all, how many workers to launch, and what context each one receives are all decided by the model at run time — and the time spent organizing counts against the same four-hour budget. Working sequentially alone is a perfectly valid workflow. Every deliverable can therefore be read next to the organization that produced it.\n\n## The scoreboard: a 36.95% ceiling and heavy specialization\n\nThe paper evaluates 9 models in 25 model-effort configurations: Claude Opus 5.5 and Sonnet 5.5 in Claude Code, GPT-6 Astra and GPT-6.1 Sol in Codex, plus GPT-6 Luna, Kimi K3.1 Pro, GLM-5.3 Flash, DeepSeek V4.1 Flash, and Doubao-Seed-Evolving.\n\nThe strongest configuration, Opus with Claude Code at max effort, fully resolves 36.95% of tasks on average, and no configuration leads every domain. Formal mathematics comes closest to saturation, with a best resolved rate of 95.83%; the ceilings in law and finance sit at just 9.09% and 18.75%. The gap between partial credit and shippable work is stark: law tasks average a continuous score of 70.7 across configurations while only 2.9% fully pass, and Opus at high effort scores 96.98 in software engineering yet resolves just 17.65% of SWE tasks — models can write a mostly-correct analysis but rarely deliver one that clears every acceptance criterion. Content and 3D deliverables are graded against rubrics by Gemini-3.8-Flash, while the other domains use programmatic verifiers. Chinese models trail visibly on this table: Kimi K3.1 Pro at 16.48%, Doubao at 10.84%, DeepSeek V4.1 Flash at 6.08%, and GLM-5.3 Flash at 4.43%. Effort sensitivity also differs: across five settings Opus swings from 22.02% to 36.95%, while Astra varies only between 27.34% and 30.80%.\n\n## Being good at the task is not being good at managing it\n\nThe key experiment swaps blueprints: on a challenging subset of 32 tasks, the execution models (Kimi, DeepSeek, Luna, Doubao) are held fixed while only the workflow author changes. Workflows written by Opus improve all four executors, adding 6.41 points on average in the static mode — Kimi climbs from 35.75 to 46.61, DeepSeek from 18.37 to 25.24. Astra's blueprints, by contrast, average −2.53 points in the static mode and drag down both Kimi and DeepSeek, even though Astra on its own (27.34% resolved at max effort) is far stronger than any of these executors. The ability to do the work well does not automatically become the ability to organize others to do it.\n\nThe data also pins down two management styles. At high effort, Opus averages 0.86 workflow calls per task with 5.55 workers each — few batches, big teams — while Astra averages 2.09 calls with just 1.72 workers, and about 99% of its workflows contain at most four workers: high frequency, small squads.\n\n## So what\n\nThis benchmark turns orchestration from marketing language into a measurable, transferable asset: the same weak models gain more than 6 points just by plugging into a better blueprint. The cost math is blunt: Opus at high effort resolves 34.11% at $6.66 per task, while Astra at xhigh spends $29.26 to resolve 30.80% — 4.4 times the cost for 3.31 points less. For anyone building agents, organization is the next battleground beyond single-model benchmarks; for anyone choosing one, ask whether you are hiring a strong worker or a competent manager.\n\nPaper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.12856","workflowarena-multi-agent-orchestration","2026-11-10T15:30:00Z","2026-11-10T15:10:36.546024Z","2026-11-10T15:10:36.546034Z",true,"agent",66,[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},"8857d393-a5f7-4b2a-8439-bf87ac324d4f","数学建模基准自曝:12个模型算对却只得49.82%","tb-mathmodel-agentic-verifiable-math-modeling-bench","2026-11-08T03:15:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"37e0486a-225d-4b44-b189-206bd756210c","多轮 Agent 强化学习提升了什么:46.8% 的真相","multi-turn-agentic-rl-budget-dependence","2026-11-06T01:18:33+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"99c1eb7b-9226-4d61-afdb-5d54f920a85c","研究品味实测:AI 智能体的想法匹配只有人类的 18%","ideataste-llm-research-taste-benchmark","2026-11-04T11:13:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"a50b0336-dabe-405e-b8c4-d27156ab6d24","60% 权重转私有测试集:开放权重落后榜首 12 分","artificial-analysis-intelligence-index-v5","2026-10-28T07:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"ba7f04d1-2fe5-4e1d-90a6-94ec3c4b70df","阿里 UserArena:64,729 条购物轨迹当考场","userarena-shopping-agent-user-simulation-benchmark","2026-10-26T09:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"aa294551-fbef-4cf4-b466-1928ab1925bb","朴素 RAG 拿下 6\u002F7 组评测:复杂记忆系统被比下去了","naive-rag-beats-complex-agentic-memory","2026-10-25T07:17:45+00:00"]