[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-cheap-proxy-tests-can-predict-costly-ai-agent-benchmarks":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},3427,"cheap-proxy-tests-can-predict-costly-ai-agent-benchmarks","Cheap Proxy Tests Can Predict Costly AI Agent Benchmarks","A new framework called PACE claims to predict how LLM agents score on expensive benchmarks using a small set of cheaper, faster tests.","Researchers say you no longer need to spend thousands of dollars running full agent benchmarks to know how a model will perform.\n\nA team published PACE, a framework that selects a compact subset of standard non-agentic tests — the kind that measure reasoning and code generation in isolation — and uses them to predict how a model will score on heavyweight agentic benchmarks like SWE-Bench and GAIA. The method fits a regression model mapping scores on a small \"proxy\" set to scores on the full target benchmark. Tested across 14 models and four agentic benchmarks, PACE-Bench hit a mean absolute error under 4%, a Spearman correlation above 0.80, and roughly 85% accuracy in ranking models against each other — all at less than 1% of the cost of running the real thing.\n\nThe gap between cheap capability tests and expensive agent evaluations has been a quiet bottleneck in AI development: labs often can't afford to run full benchmark suites continuously during training. If proxy scores hold up outside the paper's controlled conditions, teams could make faster model selection and routing decisions without burning through compute budgets. The framework also surfaces which specific skills each agentic benchmark actually demands, which is itself useful signal.\n\nThat said, a regression trained on 14 models is a thin basis for confidence — and proxy benchmarks have a well-known failure mode where the proxy becomes the target, eroding the predictive relationship over time.","[\"ai\",\"benchmarks\",\"llm\",\"research\"]","2026-07-03T04:00:00.000Z","2026-07-03T05:45:05.304Z","2026-07-03T05:45:08.203Z","published",null,[],"ai",[24,26,27,28],"benchmarks","llm","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.02032",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"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":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]