[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-openai-biobasicsbench-wet-lab-agents":3,"topics-all":39,"news-related-6fd14d23-04ef-43c0-bb9a-fc365a6a6843":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":25,"news_slug":32,"published_at":33,"created_at":34,"modified_at":35,"is_published":36,"publish_type":37,"image_url":14,"view_count":38},"6fd14d23-04ef-43c0-bb9a-fc365a6a6843","OpenAI湿实验基准:榜首66.7%,菌落还是数错","OpenAI与酶设计公司Ligo发布 BioBasicsBench:100 个真实实验任务,考智能体在凝胶、显微镜堆栈、测序读段等原始数据上读图、数菌落、查对照。7 模型 18,500 份评分里,GPT-6 Astra 以 66.7% 居首,GPT-5.2 仅 31.2%;但榜首也会选错菌落、计数多算约 12%。","胶图上的条带、平板上的菌落、显微镜的逐层堆栈——这些是湿实验室里每天都要打交道的证据。AI 智能体到底能不能替科学家把它们读对?OpenAI 交出的答卷相当不客气:在一百个日常湿实验任务上,成绩排在榜首的 GPT-6 Astra 平均也只拿到约三分之二的分数,而且照样会数错菌落。\n\n## 100 个任务,全部来自真实实验\n\nOpenAI 与专注酶设计的生物技术公司 Ligo 合作构建了 BioBasicsBench,覆盖蛋白方法、成像与流式细胞术、定量与检测、核酸方法、分子克隆五个日常板块。任务不是从教科书里挑的:科学家把自己实验里的真实数据——包括失败的实验和模糊的图像——变成考题,每道题配专家评分细则,再过三道评审:一位未参与出题的 Ligo 科学家、一位外部湿实验室 PhD 级科学家、最后一次内部终审。\n\n证据形态也故意保留实验室的\"脏\":凝胶与印迹出现在 31 个任务里,协议、笔记与仪器规格 27 个,表格与仪器导出(含流式原始 FCS 文件)20 个,台面与试剂照片 18 个,显微镜图像 14 个,仪器屏幕与曲线 13 个,测序文件 12 个,操作视频与平板菌落各 7 个,另有 21 个任务附带梯图、质粒图谱等参考材料(类别有重叠)。官方明说:现有基于论文图表的基准,恰恰缺少这种歧义和排障场景。\n\n## 七个模型排队挨考,榜首只有 66.7%\n\nOpenAI 评测了 GPT-5.2 到 GPT-6 Astra 的 7 个模型、37 组模型与推理力度设置,每组在 100 个任务上各跑 5 次,合计 18,500 份评分。xhigh 最高推理档下的平均得分(mean reward):\n\n| 模型 | 平均得分 | 每任务输出 token |\n| --- | --- | --- |\n| GPT-6 Astra | 66.7% | 4,704 |\n| GPT-6.1 Sol | 62.0% | 5,238 |\n| GPT-5.6 Sol | 43.8% | 9,099 |\n| GPT-6 Luna | 38.3% | 13,120 |\n| GPT-5.5 | 35.4% | 11,135 |\n| GPT-5.4 | 33.2% | 13,717 |\n| GPT-5.2 | 31.2% | 27,620 |\n\n两个信号值得注意。其一,GPT-6.1 Sol 与 GPT-5.6 Sol 之间隔着 18.2 个百分点的断层,OpenAI 的解读是更强的原生多模态能力可能是关键(官方口径)。其二,效率在代际间反转:GPT-5.2 每个任务烧 27,620 个输出 token 只得 31.2%,是 Astra 的约 5.9 倍——新一代模型不是靠把推理链拉长赢的。\n\n## 拿高分的模型,实验室里照样翻车\n\nOpenAI 特意展示榜首模型的三次翻车。菌落 PCR 任务里,Astra 把泳道条带分组看对了,实验归因却接反了:它默认更大的 PCR 产物代表克隆成功,选了 4、6、7、8 泳道,参考答案按对照逻辑选 5、9、10——照它选下去,下游纯化出来的是错误质粒。数菌落任务里,它的计数比参考值高出约 12%:近邻菌落粘成一个标记、可见菌落漏标、空白培养基也被数进去;而计数虚高,按固定体积与稀释度估算的转化效率会同比例虚高。还有一题,它检查了明、暗两种液泡,结论却只保留支持\"液泡更暗\"的那一半观察。\n\n翻车之外也有亮点:一道 DNA 克隆题里,GPT-5.6 Sol 看到阴性对照板没长菌落就直接下结论,Astra 注意到平板边缘标签写着 LB Kan50(卡那霉素),而实验用的是氨苄青霉素,于是把这块板排除出可靠对照;一道死细胞染色题里,GPT-5.6 Sol 凭单一染料的亮度就把细胞判死,Astra 则指出两个检测通道的窗口在 525–540 纳米重叠、亮度不足以单独定论,主动要求补对照;还有一题它认出细胞\"忽隐忽现\"其实是显微镜的 z 轴逐层扫描,不是生物学现象。\n\n## 所以呢\n\n这个基准切开的,是\"会答生物题\"和\"会做实验\"的差别:后者考的是在模糊证据面前压住过度解读、承认证据不足。66.7% 的现实含义是:日常湿实验分析还没有模型能托底,Astra 的 95% 置信区间上界也只有 73.1%。OpenAI 在文末说,持续跟踪这些能力\"将帮助我们理解其对生物安全与生物安保的影响\"——这份榜单既是能力榜,也是风险清单。对做 AI for Science 的人来说,别急着问模型能不能进实验室,先问它敢不敢承认:这张胶图说明不了问题。(原始发布:[OpenAI 官方文章](https:\u002F\u002Fopenai.com\u002Findex\u002Fmeasuring-how-ai-agents-handle-wet-lab-science))","https:\u002F\u002Fopenai.com\u002Findex\u002Fmeasuring-how-ai-agents-handle-wet-lab-science","15975962-b5fe-49e5-ae68-687ba6cb7015",[11,15,19,22],{"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":18,"color":14},"9112951a-2abb-4214-b63a-385ec7afb2ba","ai-for-science","AI for Science 专题：追踪 AI 在生命科学、化学材料、物理世界模型等科学方向的关键突破",{"id":20,"name":21,"slug":21,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":23,"name":24,"slug":24,"description":14,"color":14},"42e59a88-7795-47dc-a334-ef1e72c24347","openai",[26],{"id":27,"lang":28,"title":29,"summary":30,"content":31},"4b5da4d1-119e-48ad-bdfc-aaad55fe9d98","en","BioBasicsBench: top model 66.7%, still miscounts colonies","OpenAI's BioBasicsBench: 100 real wet-lab tasks; GPT-6 Astra leads at 66.7%, GPT-5.2 at 31.2% — yet the leader still miscounts colonies by ~12%.","Gel bands, colony plates, z-stacks of microscopy — this is the evidence wet-lab scientists wrestle with every day. Can AI agents read it correctly on their behalf? OpenAI's answer is blunt: across a hundred routine wet-lab tasks, its top-scoring model, GPT-6 Astra, earns on average only about two-thirds of the available credit — and still miscounts colonies.\n\n## 100 tasks, all from real experiments\n\nBioBasicsBench was built with Ligo, a biotech company focused on enzyme design. It covers five areas of routine wet-lab work: protein methods, imaging and cytometry, quantification and assays, nucleic acid methods, and molecular cloning. The tasks were not lifted from textbooks: scientists turned data from their own experiments — including failed runs and ambiguous images — into questions, each paired with an expert-written grading rubric and put through three reviews: a Ligo scientist who had not written the task, an external PhD-level scientist working in a wet lab, and a final internal review. Answers are scored against individual rubric criteria, so the benchmark can tell a correct conclusion apart from a partially correct analysis.\n\nThe evidence stays deliberately messy. Gels and blots appear in 31 tasks; protocols, notes and specs in 27; tables and instrument exports (including raw flow-cytometry FCS files) in 20; bench, apparatus and reagent photos in 18; microscopy in 14; instrument screens and plots in 13; sequencing files in 12; protocol videos and agar plates in 7 each; and 21 tasks carry reference sheets such as ladder charts and plasmid maps (categories overlap). OpenAI says existing benchmarks built around figures selected for papers miss exactly this kind of ambiguity and troubleshooting.\n\n## Seven models queue up; the top score is 66.7%\n\nOpenAI evaluated seven models — from GPT-5.2 to GPT-6 Astra — across 37 model and reasoning-effort settings, with five attempts on each of the 100 tasks per setting: 18,500 graded results in total. Mean reward at the highest xhigh effort:\n\n| Model | Mean reward | Output tokens \u002F task |\n| --- | --- | --- |\n| GPT-6 Astra | 66.7% | 4,704 |\n| GPT-6.1 Sol | 62.0% | 5,238 |\n| GPT-5.6 Sol | 43.8% | 9,099 |\n| GPT-6 Luna | 38.3% | 13,120 |\n| GPT-5.5 | 35.4% | 11,135 |\n| GPT-5.4 | 33.2% | 13,717 |\n| GPT-5.2 | 31.2% | 27,620 |\n\nTwo signals stand out. First, an 18.2-point cliff separates GPT-6.1 Sol from GPT-5.6 Sol; OpenAI reads this as stronger native multimodal capability being key (the vendor's own interpretation). Second, efficiency flips between generations: GPT-5.2 burns 27,620 output tokens per task for 31.2% — roughly 5.9x what Astra spends — so the new generation is not winning by writing longer reasoning chains.\n\n## The top model still faceplants in the lab\n\nOpenAI showcases three failures from its own leader. In a colony PCR task, Astra grouped the gel bands correctly but reversed the experimental attribution: assuming the larger PCR product marked successful cloning, it picked lanes 4, 6, 7 and 8, while the reference — following the controls — selects lanes 5, 9 and 10. Follow its pick and you would purify the wrong plasmid downstream. In a counting task its tally runs about 12% over the reference: close neighbours merged into one mark, visible colonies left unmarked, empty agar counted as a colony — and an inflated count inflates the transformation-efficiency estimate by the same proportion at fixed plated volume and dilution. In another task it inspected both darker and lighter vacuoles, then kept only the half of the evidence that supported its 'vacuoles are darker' conclusion.\n\nThere are genuine wins too. In a DNA-cloning task, GPT-5.6 Sol saw no colonies on the negative-control plate and concluded away; Astra noticed the rim label read LB Kan50 — kanamycin — while the experiment used ampicillin, and correctly rejected that plate as a reliable control. In a dead-cell staining task, GPT-5.6 Sol called the cells dead from a single dye's brightness, while Astra pointed out that the two detection channels overlap at 525–540 nm, that brightness alone cannot settle the question, and asked for better controls. In another task it recognized cells 'disappearing' as a z-stack depth scan rather than a biological event.\n\n## So what?\n\nThe benchmark separates answering biology questions from doing experiments: the latter is about resisting over-interpretation and admitting when the evidence is insufficient. 66.7% means no model can yet be trusted with routine wet-lab analysis — the upper bound of Astra's 95% confidence interval is 73.1%. OpenAI closes by saying that tracking these capabilities will also help us understand their implications for biosafety and biosecurity — this leaderboard is both a capability chart and a risk register. For anyone building AI for science, don't ask whether the model can enter the lab; ask whether it dares to say: this gel does not support a conclusion. (Source: [OpenAI's announcement](https:\u002F\u002Fopenai.com\u002Findex\u002Fmeasuring-how-ai-agents-handle-wet-lab-science))","openai-biobasicsbench-wet-lab-agents","2026-11-08T19:30:00Z","2026-11-08T19:08:29.269555Z","2026-11-08T19:08:29.269570Z",true,"agent",31,[40,48],{"slug":17,"tag_slug":17,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":36,"created_at":46,"modified_at":47},"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":36,"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},"158f2181-ecf3-4e5c-ac5a-16e5c45c9d9a","千篇论文建成AI实验室:前沿模型74%胜过人类作者","scientist-world-1000-research-environments","2026-11-08T21:08:04+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"ac67a21d-fd2c-48b4-b987-7de6a064b157","OpenAI 出题自己考:LifeVal 的三个空白","openai-lifeval-everyday-tasks-vs-gdpx","2026-11-03T11:22:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"9dffcd24-e154-4986-baab-be712e6e0442","GPT-5.6 Sol通关Factorio:低思考档打出火箭","gpt-5-6-sol-factorio-rocket-run","2026-11-01T13:20:38+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"87d43d7b-b32e-499d-af8a-68afc6506a2c","科研Agent大考:头名80.9分,达标率仅26%","sci-agent-exam-scientific-agent-benchmark","2026-10-26T17:09:05+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"d0b1ff09-4d6e-4657-abeb-7cbfca7a628a","克雷研究所回应 Navier-Stokes:百万美元奖金先过同行评审这关","clay-institute-navier-stokes-response","2026-09-13T21:07:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"7ed7fd97-8901-4c34-bef0-53a30d8c6316","OpenAI 的千禧年数学题答卷:88 小时 1 万个智能体,引发学界对未发表成果的伦理大讨论","openai-navier-stokes-controversy-unpublished-work","2026-09-12T05:30:00+00:00"]