[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-harden-method-makes-ai-benchmarks-harder-to-game":10,"sections":40},{"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":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},8287,"harden-method-makes-ai-benchmarks-harder-to-game","HARDEN Method Makes AI Benchmarks Harder to Game","HARDEN cut model accuracy by 22.7% on average versus standard benchmarks, and by as much as 49.9% versus weaker single-pass baselines.","A new evaluation method quietly makes AI benchmarks a lot less forgiving.\n\nResearchers built HARDEN, a constrained evolutionary search that rewrites benchmark questions into harder versions without changing the expected answer. It mutates inputs along complexity axes tailored to each domain, while built-in checks confirm the new question still means the same thing, still sounds realistic, and can still be executed correctly. The team tested it on three benchmarks - FinQA, PubMedQA, and ContractNLI - across three sizes of Qwen3.5 models, from 35 billion to 397 billion parameters. Accuracy fell by 22.7% on average compared to the standard benchmark versions, and by as much as 49.9% compared to a single pass of HARDEN's own search process, meaning repeated search finds much harder cases than one round alone.\n\nBenchmarks like these already shape how frontier models get ranked and marketed. If a fairly simple evolutionary search can knock a fifth off accuracy on average, and half in the best case, without changing what counts as a correct answer, that says the published scores were measuring something narrower than real-world difficulty. The effect shows up hardest in finance, medicine, and contract analysis - exactly the regulated, high-stakes domains where enterprises are being sold these models as reliable.\n\nA benchmark score, it turns out, is a score against one specific set of questions - not a durable measure of what a model can actually do.","[\"ai\",\"benchmarks\",\"evaluation-methods\",\"language-models\"]","2026-09-28T04:00:00.000Z","2026-09-28T21:57:28.730Z","2026-09-28T21:57:36.012Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The dek's 'cut AI benchmark accuracy by up to half' conflates two different figures from the source — the 22.7% average drop is versus the standard benchmark, while the 49.9% figure is versus a weaker single-pass baseline of HARDEN itself, not the original benchmark accuracy — so the dek should be revised to keep those two comparisons distinct.","resolved","ai",[30,32,33,34],"benchmarks","evaluation-methods","language-models",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.30571",0,{"sections":41},[42,46,51,56,61,66,71,76,81,86,91,96,101,106],{"name":43,"slug":30,"count":44,"latest_published_at":45},"AI",4900,"2026-09-28T17:44:43.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Security","security",766,"2026-09-28T15:35:23.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Policy","policy",405,"2026-09-28T17:00:51.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Deals","deals",269,"2026-09-28T17:42:46.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",191,"2026-09-28T15:45:00.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Consumer Tech","consumer-tech",139,"2026-09-28T17:09:47.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",87,"2026-09-28T16:11:42.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Startups","startups",80,"2026-09-28T17:50:28.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":107,"slug":108,"count":109,"latest_published_at":110},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]