[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-how-big-is-that-ai-model-researchers-have-a-guess":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},4223,"how-big-is-that-ai-model-researchers-have-a-guess","How Big Is That AI Model? Researchers Have a Guess","A new benchmark uses factual recall as a proxy for parameter count, giving outsiders a rough tool to size up closed-source frontier models.","Researchers think they can estimate how large a closed-source AI model is by quizzing it on obscure facts.\n\nA team behind a paper called Incompressible Knowledge Probes (IKPs) built a benchmark of 1,400 factual questions sorted into seven tiers of obscurity. The logic: storing facts requires weights, so how many facts a model knows sets a floor on how many parameters it must have. They calibrated a log-linear formula against 93 open-weight models ranging from 135 million to 1.6 trillion parameters across 19 vendors, landing on an R-squared of 0.91. Cross-validation showed a median error of 1.48x, with 72% of estimates falling within 2x of the true count. They then applied the instrument to 201 models from 27 vendors and published estimates for every major proprietary frontier model as prediction bands rather than point estimates.\n\nClosed-source labs like OpenAI and Google have made parameter counts a trade secret, leaving the rest of the industry guessing at the true scale and cost of frontier systems. A coarse-but-calibrated external gauge matters because model size shapes everything from inference pricing to energy consumption to competitive positioning — and right now labs get to set the terms of that conversation entirely. IKPs do not end that asymmetry, but they shrink it.\n\nThe instrument is deliberately blunt: the 90% prediction interval spans roughly 3x in either direction, which the authors freely admit is too wide to settle infrastructure pricing debates. For heavily safety-tuned models it also yields a lower bound, not an estimate, because refusal policy can suppress tens of percentage points of answerable capacity. Still, order-of-magnitude rankings produced by an independent method are more than the industry has offered voluntarily.","[\"ai\",\"llm\",\"benchmarks\",\"open-source\"]","2026-07-07T04:00:00.000Z","2026-07-07T20:22:06.347Z","2026-07-07T20:22:09.236Z","published",null,[],"ai",[24,26,27,28],"llm","benchmarks","open-source",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.24827",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]