[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-minimax-h3-physical-reasoning-eval":3,"topics-all":39,"news-related-371dd90f-e8bb-46a1-967a-346c28b393df":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},"371dd90f-e8bb-46a1-967a-346c28b393df","MiniMax-H3 会不会推理物理世界?517 题实测只对 42%","新加坡国立大学、复旦大学与腾讯的联合团队给 MiniMax-H3 出了一份 517 题的物理世界推理评测:总成功率 41.97%,视频决策题最好(56%),音频歧义消解最差(27.4%)。会生成,不等于听得懂声音。","MiniMax-H3 是那种\"什么都能生成\"的模型:文本、图像、视频、音频在同一个共享潜空间框架里联合建模。上线不到两个月,它已经被蒸馏过(FastH3)、被后训练过(fal 的 H3 Max)、权重也被第三方拆解过。但一个更根本的问题一直没人系统回答:它真的能对物理世界做推理吗?新加坡国立大学、复旦大学与腾讯的 13 人研究团队最近交出了一份 517 题的\"体检报告\"(arXiv:2609.18323),结论不太客气——总成功率 41.97%。\n\n## 这套评测的刁钻之处:单模态只给一半证据\n\n按论文自述,现有视频生成评测大多有一个通病:提示词和目标内容高度重合,模型基本在\"看图说话\"。这套评测反着来:文字提示只说明要执行的操作,任务相关的关键信息被故意留白,必须从图像、音频、视频片段这些互补证据里自己推断。比如音频题会给出一张含多个候选声源的图,模型要先把声音和正确的物体对上,再把这种理解画进生成结果里。\n\n四个场景覆盖 29 个子类:\n\n- **多视角空间推理(MSR)**:单图输入,200 题,考察跨互补视角的空间关系——43.50%\n- **音频歧义消解(ADR)**:图 + 音频,146 题,用声学线索消解视觉歧义——27.40%,全场最低\n- **视频决策推理(VDR)**:前缀视频续写,100 题,对观察到的动态做出合理响应——56.00%,全场最高\n- **视听整合推理(AVIR)**:视频 + 音频,71 题,把听觉证据或口头约束融进续写与编辑——47.89%\n\n评分不是对答案:每条生成结果由 3 名专家独立判断是否满足语义约束,允许存在多个合法的视觉呈现。总成功率按全部 517 题计算,不是四个场景的简单平均。\n\n## 42 分背后:会生成,不等于听得懂\n\n两个发现值得展开。第一,**音频接地是最短的板**。论文列出的典型失败包括:把声音关联到错误的可见声源、混淆相似的机械声。也就是说,画面越\"看起来对\",这种错位越难被察觉——评测框架的 README 说得很直白:视觉合理性是不够的,一段视频可以看起来很有说服力,却不满足任务要求的空间、时序或视听条件。第二,视频题(56%)明显好于音频题(27.4%),团队自己的结论是:**提供多模态输入,并不保证任务被可靠完成**,有效的多模态整合仍是未解问题。\n\n论文也提前打好了预防针:四个场景的数据、提示词和生成目标各不相同,这些成功率描述的是任务级表现,不能当作不同模态\"内在价值\"的对照实验;而且生成结果只是行为证据,失败可能出在感知、证据整合或生成的任何一环。\n\n## 泼两盆冷水,再说一句公道话\n\n冷水一:目前评测数据集、评测结果与自动评测管线在仓库里都还是未勾选的 TODO(团队称正在构建一条 DeepSeek 驱动的自动评测管线),所有数字都是单一团队的自报结果,第三方暂时无法复现。冷水二:它只测了 MiniMax-H3 一个模型,\"omni 模型普遍如此\"是推论,不是结论。\n\n公道话是:这份评测提出的问题比给出的数字更有价值。当下 omni 生成模型的公开竞赛集中在画质、时长、分辨率这些可见指标上,而\"是否真正整合了跨模态证据\"这个维度,此前少有评测框架专门去碰。这篇论文在 9 月 18 日的 Hugging Face Daily Papers 榜排到第二(92 个赞),说明社区确实在渴这个方向。对做视频生成的团队,音频-视觉对齐大概是被低估得最狠的一块短板。\n\n所以,下次再看到\"一个模型生成整个声场\"的演示,不妨多问一句:它生成出来了,但它听懂了吗?\n\n参考:arXiv:2609.18323(arxiv.org\u002Fabs\u002F2609.18323)· 评测框架 github.com\u002Fgulucaptain\u002FMiniMax-H3-Reason","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.18323","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,19,22],{"id":12,"name":13,"slug":13,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",null,{"id":16,"name":17,"slug":17,"description":18,"color":14},"f3be854a-50f0-411c-893d-16d0df6def02","h3-series","MiniMax H3 专题：持续追踪 H3 的发布、开源、蒸馏与部署全链路",{"id":20,"name":21,"slug":21,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":23,"name":24,"slug":24,"description":14,"color":14},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[26],{"id":27,"lang":28,"title":29,"summary":30,"content":31},"4bcb6f4c-d1bb-4345-aa15-4f9d035375f7","en","MiniMax-H3 Physical-Reasoning Test: 42% Overall, Audio Just 27%","NUS-led team tested MiniMax-H3 on 517 multimodal tasks: 41.97% overall, video best at 56%, audio worst at 27.4%.","MiniMax-H3 is one of those \"generates everything\" models: text, images, video and audio jointly modeled in a shared latent framework. Since its open-weight release in late July it has been distilled (FastH3), post-trained (fal's H3 Max) and torn down by third parties. But a more fundamental question had no systematic answer: can it actually reason about the physical world? A 13-author team from the National University of Singapore, Fudan University and Tencent just published a 517-instance \"physical exam\" (arXiv:2609.18323), and the verdict is blunt — an overall success rate of 41.97%.\n\n## What makes this eval nasty: every single modality gives only half the evidence\n\nBy the paper's own account, most existing video-generation benchmarks share a flaw: prompts closely match the target content, so the model is mostly doing caption-to-video. This eval inverts that. The text prompt only specifies the operation to perform; task-relevant information is deliberately left unstated and must be recovered from complementary image, audio and video evidence. In the audio scenario, an image may contain several candidate sound sources — the model must first bind the supplied sound to the correct object, then express that interpretation in the generated output.\n\nFour scenarios, 29 subcategories:\n\n- **Multi-view Spatial Reasoning (MSR)** — single image, 200 instances, spatial relationships across complementary views: 43.50%\n- **Audio-based Disambiguation Reasoning (ADR)** — image + audio, 146 instances, resolve visual ambiguity via acoustic cues: 27.40%, the floor of the whole eval\n- **Video-based Decision Reasoning (VDR)** — prefix-video continuation, 100 instances, respond appropriately to observed dynamics: 56.00%, the ceiling\n- **Audiovisual Integrated Reasoning (AVIR)** — video + audio, 71 instances, fold auditory evidence or spoken constraints into continuation and editing: 47.89%\n\nScoring is not answer matching: three experts independently judge whether each output satisfies the intended semantic constraint, and multiple valid visual realizations are allowed. The overall rate is computed over all 517 instances, not averaged across scenarios.\n\n## Behind the 42%: generating well does not mean listening well\n\nTwo findings stand out. First, **audio grounding is the shortest plank**. The listed failure modes include associating a sound with the wrong visible source and confusing similar mechanical sounds. The framework's README states it plainly: visual plausibility is insufficient — a video can look convincing while failing the required spatial, temporal or audiovisual condition. Second, video tasks (56%) clearly beat audio tasks (27.4%); the team's own conclusion is that **supporting multimodal inputs does not ensure reliable task completion**, and effective multimodal integration remains an open problem.\n\nThe paper also pre-empts over-reading: the four scenarios differ in data, prompts and generation targets, so these rates describe task-level performance, not a controlled comparison of the intrinsic value of modalities; and generated outputs are behavioral evidence — a failure could originate in perception, evidence integration, or generation.\n\n## Two buckets of cold water, then one fair point\n\nCold water #1: the evaluation dataset, the results, and the auto-evaluation pipeline are all still unchecked TODOs in the repo (the team says it is building a DeepSeek-powered automated pipeline), so every number is a single-team self-report with no third-party replication yet. Cold water #2: only MiniMax-H3 was tested — \"omni models in general are like this\" is an extrapolation, not a finding.\n\nThe fair point: the question this eval raises is more valuable than its numbers. The public omni-generation race has concentrated on visible metrics — resolution, duration, visual quality — while the paper argues existing frameworks rarely exploit omni-modal inputs at all. It ranked #2 on Hugging Face's Daily Papers board for Sep 18 (92 upvotes), so the community is clearly hungry for this direction. For video-generation teams, audio-visual alignment may be the most underestimated weak spot.\n\nSo the next time you see a demo claiming \"one model generates the whole soundscape,\" ask one more question: it generated — but did it understand what it heard?\n\nReferences: paper [arXiv:2609.18323](https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.18323) · framework [GitHub: gulucaptain\u002FMiniMax-H3-Reason](https:\u002F\u002Fgithub.com\u002Fgulucaptain\u002FMiniMax-H3-Reason)","minimax-h3-physical-reasoning-eval","2026-09-19T17:11:04Z","2026-09-19T17:11:08.017349Z","2026-09-19T17:11:08.017360Z",true,"agent",61,[40,49],{"slug":41,"tag_slug":41,"title_zh":42,"title_en":43,"intro_zh":44,"intro_en":45,"id":46,"is_active":36,"created_at":47,"modified_at":48},"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":17,"tag_slug":17,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":36,"created_at":55,"modified_at":56},"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},"1bf979b5-2b52-4a19-8e2d-e0a549bb24ba","fal 后训练版 MiniMax H3:5 秒视频约 3 秒生成,吞吐 35 倍","fal-h3-max-post-trained-video","2026-08-29T23:02:12+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"b95b93e8-294a-4c5b-b53d-ce6ea07c1519","SemComp-Bench 登顶 Hugging Face 日榜:视频生成开始考「任务做没做成」","semcomp-bench-video-task-completion","2026-08-20T13:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"aad00b18-d354-48b5-ad21-62b53150b8c6","MiniMax H3 开源实测:你下载的权重,和 API 里跑的不是同一个模型","minimax-h3-local-vs-api-gap","2026-08-15T17:07:24+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"6e3002da-c1fd-4a6d-b903-4f65b976dd04","MiniMax H3 首个商用落点：美图 RoboNeo 接入背后,通用多模态模型的\"可编辑性\"才刚开始被检验","roboneo-minimax-h3-multimodal-editing","2026-08-03T18:02:02+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"6f375936-79af-4622-a75e-d802ade563e0","MiniMax H3 不只是 2K 视频：它想把生成、参考和编辑收回一个模型","minimax-h3-omnimodal-video-unified-generation-editing","2026-08-03T04:08:31+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"a970be54-0603-4446-b92e-67c887c0788a","ComfyUI Day-0 跑通 MiniMax H3:把 2K 多模态视频生成压到 RTX 3060","comfyui-day0-minimax-h3-2k-video","2026-08-03T04:00:00+00:00"]