[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-best-ai-model-solves-only-337-of-olympiad-physics-problems":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},8110,"best-ai-model-solves-only-337-of-olympiad-physics-problems","Best AI Model Solves Only 33.7% of Olympiad Physics Problems","A new 11,586-problem benchmark called PhysElite tests 18 multimodal models on Olympiad-level physics, and even the top performer barely tops a third correct.","A new benchmark just handed AI models a physics olympiad, and most of them failed.\n\nResearchers built PhysElite, a bilingual Chinese-English benchmark made up of 11,586 Olympiad-tier physics problems. Each problem comes with a visual diagram, a step-by-step solution derivation, and a final answer. The team tested 18 open-source and closed-source multimodal language models against the set. The strongest model answered only 33.7% of the problems correctly, and the researchers also ran a step-level evaluation to see exactly where in the reasoning chain models go wrong, not just whether they land on the right final number.\n\nMost existing physics benchmarks for AI are either too easy or too narrow to test real multi-step reasoning, so headline accuracy scores can overstate how well these models actually work through diagrams and derivations. PhysElite's scale and step-by-step grading make it harder for a model to get credit by guessing a plausible final answer without doing the underlying physics. A sub-34% top score is a useful check on any claim that these systems have caught up to serious science reasoning.\n\nFor tools regularly pitched as research assistants, getting roughly two out of three olympiad physics problems wrong is worth remembering next time a demo shows a chatbot deriving equations with a flourish.","[\"ai\",\"llm-benchmarks\",\"physics\",\"multimodal-ai\"]","2026-09-28T04:00:00.000Z","2026-09-28T09:54:09.282Z","2026-09-28T09:54:20.796Z","published",null,[],"ai",[24,26,27,28],"llm-benchmarks","physics","multimodal-ai",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2608.25097",0,{"sections":35},[36,39,43,48,53,58,62,67,72,77,82,87,91,96],{"name":37,"slug":24,"count":38,"latest_published_at":18},"AI",4791,{"name":40,"slug":41,"count":42,"latest_published_at":18},"Security","security",762,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",399,"2026-09-27T18:39:02.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",261,"2026-09-27T15:30:35.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",188,"2026-09-27T20:46:36.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",151,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",135,"2026-09-26T14:30:00.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":73,"slug":74,"count":75,"latest_published_at":76},"Dev Tools","dev-tools",84,"2026-09-26T04:20:58.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Startups","startups",76,"2026-09-25T18:33:59.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":88,"slug":89,"count":85,"latest_published_at":90},"General","general","2026-09-26T17:02:42.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",30,"2026-09-24T20:07:31.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]