[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-memdoor-team-solve-qwen-gvs5h":3,"topics-all":38,"news-related-f2d8a55a-d6e2-4b2a-9b2d-a8992cf3bfde":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},"f2d8a55a-d6e2-4b2a-9b2d-a8992cf3bfde","Memdoor五Qwen协作:十题9.8压过Fable 5","开源框架Memdoor的team-solve工作流复现GVS5H方法：五个Qwen3.8-27B实例各任规划、构思、求解等角色，冷启动、只靠磁盘文件通信。在LiveCodeBench Hard前十题五轮平均解出9.8题，Fable 5单模型为8.0。五十次运行全量公开，共花35.79美元；作者声明无法排除训练污染。","本周，开源项目 Memdoor 在 GitHub 上公开了一组对照结果：五个 Qwen3.8-27B 实例组成的「小队」，在 LiveCodeBench Hard 的前十道题上五轮平均解出 9.8 题；而 Claude Fable 5 单模型一次性作答，成绩是 8.0。27B 的开放权重模型，靠编排追上了前沿闭源单模型——至少在这十道题上。\n\n## 复现别人的测试，还高 0.2 分\n\n这组数字的特殊之处在于它不是自吹自擂的跑分。Memdoor 复现的是 Lee（2026）提出的 GVS5H 方法：用原作者一模一样的提示词和循环，跑原作者题目清单的前十题，每题五轮。重建出的 harness 平均 9.8 分，略高于原作者自己 harness 的 9.6，明显高于 Fable 5 单次调用的 8.0。作为参照，GVS5H 作者在完整 100 题上测得 92.4 对 90.4——差距随规模缩小，但方向没有反转。全部 50 次运行都随仓库公开：每一次请求、回复、程序和评分器输出都可以逐条检查。\n\n## 五个角色，零共享上下文\n\nteam-solve 工作流的机制比分数更值得看。五个 Qwen3.8-27B 实例分别扮演规划者、两个并行的构思者、求解者和管理者，外加一个修复工人。每个角色都从全新上下文冷启动，彼此之间唯一的通信渠道是磁盘上的文件：规划者写计划文件，求解者写 solution.py，管理者读测试日志，修复工人打补丁。成败由一个跑样例的 shell 脚本判定，而不是模型自评——模型没有机会说「我觉得对了」。\n\n## 35.79 美元、1530 万 token 与一句污染声明\n\n成本也摆得上台面：五十次运行共花 35.79 美元、消耗 1530 万输出 token，中位数一次 40 分钟、0.39 美元（Qwen3.8-27B，128K 回复上限）。作者同时把丑话说在前面：题目来自 2025 年 3 月 29 日至 4 月 6 日的比赛，模型可能在训练中见过，「污染无法排除」。工作流本身用 YAML 编写、不绑定任何模型——指向 API 模型可以，用 Ollama、llama.cpp、vLLM 或 LM Studio 跑本地开放权重也可以，不需要任何 API key。\n\n## 编排换能力，这笔账怎么算\n\n我的看法：这不是「27B 干翻前沿模型」的胜利宣言，而是「编排 vs 单模型」争论里又一个干净的数据点。拉长到完整 100 题，92.4 对 90.4 说明前沿模型并没有被真正甩开；40 分钟一题的中位耗时，也提醒多智能体方案的时间成本常被忽略。真正有价值的是方法论纪律：冷启动加文件通信，天然杜绝了 agent 之间的上下文互污和集体跑偏；全量公开的 50 次运行，把「跑分可复现」从口号变成了可审计的工程。做 Agent 工程的团队，值得把 [team-solve 示例](https:\u002F\u002Fgithub.com\u002Fguregodevo\u002Fmemdoor\u002Ftree\u002Fmain\u002Fexamples\u002Fworkflows\u002Fteam-solve) 和 [Memdoor 仓库](https:\u002F\u002Fgithub.com\u002Fguregodevo\u002Fmemdoor) 当参考实现通读一遍。下一个问题几乎自己跳出来：同样的纪律，换 14B 甚至 7B 的模型，还能守住几分？","https:\u002F\u002Fgithub.com\u002Fguregodevo\u002Fmemdoor\u002Ftree\u002Fmain\u002Fexamples\u002Fworkflows\u002Fteam-solve","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"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},"c187600e-804c-4697-b828-1e4330e0eb10","qwen",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"cdced2ae-4da8-4e16-b7a3-fa3f94017b8e","en","Five Qwen3.8-27B Agents Score 9.8\u002F10 on LiveCodeBench Hard","Memdoor's team-solve: five Qwen3.8-27B agents sharing files on disk scored 9.8\u002F10 on ten LiveCodeBench Hard problems; Fable 5 alone posted 8.0.","Five fresh-context Qwen3.8-27B agents, one shared disk, and a shell script that decides who is right: that is the whole trick behind Memdoor's team-solve workflow, which this week posted a mean of 9.8 out of 10 on the first ten LiveCodeBench Hard problems. Claude Fable 5, answering the same ten in a single call, posted 8.0. The numbers come from the project's GitHub repository, with all fifty runs published in full.\n\n## A replication that scored above the original\n\nWhat makes the result unusual is that it is a replication, not a self-run benchmark. Memdoor rebuilt the GVS5H method by Lee (2026) — using the original authors' exact prompts and loop, on the first ten problems of their list, five passes each. The rebuilt harness averaged 9.8 of 10, ahead of the 9.6 the original authors measured for their own harness, and well clear of Fable 5's 8.0 in a single call. Across the full 100-problem list the original authors report 92.4 versus 90.4, so the gap narrows with scale — but the direction holds. Every one of the 50 runs is published: each request, reply, program and grader output can be inspected line by line.\n\n## Five roles, zero shared context\n\nThe mechanics matter more than the score. Five Qwen3.8-27B instances play planner, two parallel ideators, solver and manager, with a repair worker on top. Each role starts from a fresh context; the only channel between them is files on disk. The planner writes a plan file, the solver writes solution.py, the manager reads the test logs, the repair worker patches. A shell script that runs the samples decides success — not a model grading itself, so no agent gets to say \"I think this is correct.\"\n\n## $35.79, 15.3M tokens and one contamination caveat\n\nThe economics are published too: fifty runs cost $35.79 — counting 16 incomplete attempts, $38.77 — and consumed 15.3 million output tokens; a median run took forty minutes and $0.39 on Qwen3.8-27B with a 128K reply cap. The authors flag contamination up front: the contests date from 29 March to 6 April 2025, and the model may have seen them in training, so contamination cannot be ruled out. The workflow itself is written in YAML and names no model — point it at an API model, or run it on local open weights through Ollama, llama.cpp, vLLM or LM Studio, with no key required.\n\n## Orchestration versus one big model\n\nMy read: this is not a victory lap for 27B models over the frontier. It is one more clean data point in the orchestration-versus-scaling argument. Stretched to the full hundred problems, 92.4 against 90.4 says the frontier model is not actually beaten; the forty-minute median run time is the quiet cost that multi-agent setups keep ignoring. What is genuinely valuable is the discipline: cold starts plus file-only communication rule out cross-contamination between agents, and publishing all fifty runs turns \"replicable benchmark\" from slogan into auditable engineering. Teams building agent pipelines should read the [team-solve example](https:\u002F\u002Fgithub.com\u002Fguregodevo\u002Fmemdoor\u002Ftree\u002Fmain\u002Fexamples\u002Fworkflows\u002Fteam-solve) and the [Memdoor repository](https:\u002F\u002Fgithub.com\u002Fguregodevo\u002Fmemdoor) as a reference implementation. The next question almost asks itself: with the same discipline, how far can the model shrink — 14B, 7B — before the score falls apart?","memdoor-team-solve-qwen-gvs5h","2026-11-01T21:09:00Z","2026-11-01T21:09:11.813479Z","2026-11-01T21:09:11.813492Z",true,"agent",227,[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},"278dc1cc-ab13-4de3-9ed4-27f6732dda3c","五个27B组队靠账本编排追平Fable 5,复现只承认打平","gvs5h-ledger-orchestration-qwen-fable-rerun","2026-11-01T09:20:28+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"a9ca0a44-b4c9-482a-8e0f-4a92295dfd74","Baseten双模型登顶决策榜:不写字,只打分","baseten-mammoth-elephant-decision-models","2026-10-26T23:15:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"206ea36a-1eea-463f-a241-1e3b32f5ec2d","USTC GraphForge:证据图把任务和 rubric 钉在一起,Qwen3.6-27B 涨三基准","graphforge-ustc-qwen36-27b-evidence-graph","2026-10-04T03:05:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"448dbda7-e4ac-44ba-8a40-7af01012b8df","把仓库\"换脸\"再测:SWE-bench 高分最多掉 14.4 分,编码 agent 疑在背题","schrodinger-repo-swe-bench-memorization","2026-09-24T17:11:54+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"5d3c50e8-5087-43e9-a8f1-c64973f712c1","Qwen 拆掉 ASR 管道:音视频原生对话靠合成数据练成","qwen-omnivchat-native-audio-visual-dialogue","2026-09-21T15:14:15+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"dcb1506b-87fa-422f-89e0-bb62afcc2b4c","BenchLM 8 月榜:Qwen3.8 Max 79.2 分领跑开源 LLM,MiniMax M3 跻身三强","qwen3-8-max-benchlm-aug-2026","2026-08-28T06:00:00+00:00"]