[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-agentrouter-slashes-ai-agent-inference-costs-by-72-percent":10,"sections":35},{"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":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},7283,"agentrouter-slashes-ai-agent-inference-costs-by-72-percent","AgentRouter Slashes AI Agent Inference Costs by 72 Percent","A 12M-parameter classifier routes each agent workflow step to the right AI model, cutting inference costs 72 percent with under 3 percent quality loss.","A new routing system decides, step by step, which AI model should handle each part of an agent's task - and claims to cut costs by 72 percent without gutting quality.\n\nResearchers built AgentRouter, a 12-million-parameter classifier that assigns each step in a multi-step AI agent workflow to one of four model tiers, from lightweight models up to frontier-class systems, using five features available at routing time. It adds less than 5 milliseconds of overhead per step on an A100 GPU. Trained on 50,000 annotated agent trajectory steps covering planning, coding, research, and data-analysis tasks, it hits 91 percent routing accuracy on the simplest steps and 76-82 percent on the hardest, frontier-tier ones. Overall it retains 97.3 percent of the task-completion quality of an all-frontier-model baseline while cutting inference costs 72 percent.\n\nMost enterprise agent systems still send every step - including trivial formatting or lookup work - to the priciest model on hand, because existing routers like RouteLLM and FrugalGPT were built to assign single-turn queries, not multi-step trajectories. Applied step by step on the same benchmarks, those two tools manage only 31 percent and 44 percent cost reductions, because they miss how much complexity swings within one task. That gap is the real finding here: as agentic workflows spread through enterprises, the routing layer, not the model itself, may be where the cost savings actually get won.\n\nIt is also a tidy rebuttal to the assumption that bigger is always safer - sometimes a formatting step just needs a formatting-sized model.","[\"ai-agents\",\"model-routing\",\"inference-costs\",\"llm-research\"]","2026-09-23T04:00:00.000Z","2026-09-23T05:00:53.530Z","2026-09-23T05:00:58.756Z","published",null,[],"ai",[26,27,28,29],"ai-agents","model-routing","inference-costs","llm-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.22951",0,{"sections":36},[37,40,44,49,54,59,63,68,73,78,83,88,93,98],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",4249,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",706,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",369,"2026-09-23T02:13:52.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Deals","deals",202,"2026-09-22T23:00:04.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",168,"2026-09-22T23:56:03.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",133,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",110,"2026-09-22T20:00:00.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Software","software",80,"2026-09-22T23:32:52.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Dev Tools","dev-tools",79,"2026-09-22T22:21:13.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",65,"2026-09-22T22:06:48.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",45,"2026-09-22T15:35:06.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",43,"2026-09-21T23:48:56.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",27,"2026-09-22T13:00:00.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]