[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-st-bench-shows-multi-agent-ai-wins-on-science-data-at-a-cost":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},10591,"st-bench-shows-multi-agent-ai-wins-on-science-data-at-a-cost","ST-Bench Shows Multi-Agent AI Wins on Science Data, at a Cost","A new benchmark finds AI agent teams beat single models on scientific data tasks, but the edge is coverage, not accuracy, and costs more compute.","Teams of AI agents can out-analyze a single model on real scientific data, a new benchmark finds - but only by casting a wider net, not by being smarter.\n\nResearchers built ST-Bench, a test set of 100 data-science tasks drawn from published Earth-science research in hydrology, agriculture, and wetland methane studies, expanded into 2,067 queries and checked by domain experts. They ran five multi-agent-system generation methods under two training setups against single-agent baselines, all built on the same GPT-5 backbone. Nine of ten multi-agent configurations beat the cheapest single-agent baseline, and the best one scored nearly three times higher on the benchmark's composite measure. But that advantage mostly came from the multi-agent systems producing usable numerical answers on more queries, not from those answers being more accurate.\n\nThat's a meaningful distinction for anyone building research-automation tools: multi-agent setups look great in the aggregate stats, but the gain is about finishing more of the job, not doing each piece better. And it isn't free. The strongest configuration took about four times as long as a single agent to run, though a cheaper setup captured most of the benefit for under twice the cost.\n\nCoding and math benchmarks made multi-agent collaboration look like a near-universal win; ST-Bench is a reminder that messier, real-world analysis tasks reward brute-force coverage more than teamwork, and that extra agents still show up on the compute bill.","[\"ai\",\"multi-agent-systems\",\"benchmarks\",\"research-automation\"]","2026-10-07T04:00:00.000Z","2026-10-08T21:33:01.167Z","2026-10-08T21:33:06.512Z","published",null,[],"ai",[24,26,27,28],"multi-agent-systems","benchmarks","research-automation",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.07763",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",6448,"2026-10-07T18:45:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",904,"2026-10-07T19:53:42.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",474,"2026-10-07T18:23:21.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",453,"2026-10-07T23:58:31.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",222,"2026-10-07T21:19:54.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Science","science",186,"2026-10-06T21:20:39.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",174,"2026-10-07T17:41:41.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",113,"2026-10-07T18:10:00.000Z",{"name":76,"slug":77,"count":73,"latest_published_at":78},"Startups","startups","2026-10-07T23:36:57.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Dev Tools","dev-tools",105,"2026-10-07T16:59:11.000Z",{"name":85,"slug":86,"count":87,"latest_published_at":88},"General","general",61,"2026-10-07T22:00:24.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"Gaming","gaming",56,"2026-10-07T12:00:00.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",33,"2026-10-05T11:57:17.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]