[{"data":1,"prerenderedAt":276},["ShallowReactive",2],{"project-49063454":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":28,"sentiment":29,"stance_mix":30,"scored_comments":34,"aspects":35,"strip":42,"post_text":45,"what":46,"said":81,"maker_in_thread":149,"shape":159,"grid":163,"attempts":238,"same_week":239,"maker_shows":183,"maker_hidden":268,"reader_version":269,"card":270},"49063454","world-model-optimizer","Optimize and serve models with Fable quality at half the cost","title","Show HN: Optimize and serve models with Fable quality at half the cost","https:\u002F\u002Fgithub.com\u002Fexperientiallabs\u002Fworld-model-optimizer","github.com","SilenN",1785108915,2026,71,30,937,4366,22,"notable","ai","model routing and distillation","cli",[24,25,26,27],"model routing","model distillation","openai compatible endpoint","agent traces","debated",0.034,{"positive":31,"neutral":32,"mixed":33,"critical":31},4,6,0,14,{"idea":36,"pricing":38,"docs":39,"privacy":40,"performance":41},{"positive":37,"critical":37},1,{"positive":37,"critical":33},{"positive":33,"critical":37},{"positive":33,"critical":33},{"positive":33,"critical":37},[43,33,44,33,44,33,43,33,44,44,33,44,33,37,33,43,33,44,33,43,33,37,37,33,33,33,33,33,37,33],5,3,"Hi HN, we built world-model-optimizer, an open-source tool for continually improve models specialized to agents. Today we are launching `wmo serve`, a tool to route repetitive tasks to distilled smaller models.\n\nAgent traces you already capture are opportunities to get signal on how to make your model cheaper, faster, better. We continuously improve\n- your specialized model through distillation from open source models\n- model routing to frontier + custom models\n- token compaction to remove noise and save tokens\n\nDemo: https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=2_m4Ze6mdko\n\nPass in traces and an OpenRoute",{"problem":47,"audience":53,"pricing":58,"motivation":62,"time_to_build":66,"stack":67,"openness":78,"stage":79,"team_size":80},{"value":48,"source":49,"evidence":50},"Reduce cost of running agent models by routing repetitive tasks to smaller distilled models.","maker",[51],{"id":52},"post",{"value":54,"source":55,"evidence":56},"Developers running agents with frontier model APIs.","inferred",[57],{"id":52},{"value":59,"source":49,"evidence":60},"Hosted solution offered at a 40%+ lower cost; waitlist only.",[61],{"id":52},{"value":63,"source":49,"evidence":64},"Make agent models cheaper, faster and better using existing agent traces.",[65],{"id":52},null,[68,74],{"value":69,"source":49,"evidence":70},"Tinker (model training)",[71],{"id":72,"by":11,"text":73},"49063769","Open source models. wmo routes requests between frontier models and open source models that continuously train using Tinker. As the smaller models improve, more traffic gets…",{"value":75,"source":49,"evidence":76},"OpenRouter",[77],{"id":52},"open_source","prototype","unknown",{"praise":82,"objections":91,"requests":117,"questions":122,"alternatives":144},[83],{"value":84,"source":85,"evidence":86},"Useful idea for tuning local models.","commenter",[87],{"id":88,"by":89,"text":90},"49063682","adrianco","Local models need to be tuned to work well so this looks useful. Seems to be for general purpose model serving. I’ve been using https:\u002F\u002Fgithub.com\u002Fadrianco\u002Fretort to run…",[92,103,110],{"value":93,"source":85,"evidence":94},"No benchmarks or concrete evidence of quality yet.",[95,99],{"id":96,"by":97,"text":98},"49063733","Art9681","The absolute best way to prove this works is by releasing a model that was fine-tuned with this method and then showing benchmarks depicting the improvement delta between the…",{"id":100,"by":101,"text":102},"49063758","Reubend","Yeah, this is just slop. No benchmarks, no concrete case studies, just some vibecoded \"platform\" to finetune models on your own traces. Which is an idea that has some value, but…",{"value":104,"source":85,"evidence":105},"Title misleading; it is routing rather than distillation.",[106],{"id":107,"by":108,"text":109},"49063801","surround","The title is misleading. This is model routing, not distillation.",{"value":111,"source":85,"evidence":112},"Waitlist against Show HN rules.",[113],{"id":114,"by":115,"text":116},"49063784","dang","Ok, I think it is in your interest to wait until you have more to show, and we'll be happy to help you with reposting it once it's ready. Waitlists are against the Show HN rules…",[118],{"value":119,"source":85,"evidence":120},"Release a fine-tuned model with benchmark improvements.",[121],{"id":96,"by":97,"text":98},[123,130,137],{"value":124,"source":85,"evidence":125},"How is privacy guaranteed?",[126],{"id":127,"by":128,"text":129},"49063671","yiyingzhang","Cool idea! How do you guarantee privacy?",{"value":131,"source":85,"evidence":132},"Cost of training and using Tinker models.",[133],{"id":134,"by":135,"text":136},"49115190","handfuloflight","What are the costs to train and use the Tinker models?",{"value":138,"source":85,"evidence":139},"Cold-start requirements for new agents.",[140],{"id":141,"by":142,"text":143},"49092402","anshad2u","Interesting approach. What does the cold-start phase look like for a new agent? How many traces or runs do you typically need before the router has enough signal to safely…",[145],{"value":146,"source":85,"evidence":147},"retort (multi-language coding model experiments)",[148],{"id":88,"by":89,"text":90},{"comments":150,"answered":43,"key":151},13,[152,155,156],{"id":153,"text":154},"49063728","That's cool, thanks for sharing!",{"id":72,"text":73},{"id":157,"text":158},"49115727","Expensive, in the thousands. We have our own infra in house and are working on bringing these costs down",{"commenters":160,"max_depth":32,"maker_replies":150,"first_comment_mins":161,"lifespan_hours":162},16,28,115,[164,166,168,172,174,175,176,179,181,185,188,191,192,194,196,199,203,205,207,209,213,215,218,220,222,225,229,231,233,235],[88,33,43,165,89,33],"project\u002Fpositive\u002Fidea",[153,33,33,167,11,37],"commenter\u002Fneutral\u002Fnone",[169,37,44,170,171,33],"49063654","project\u002Fneutral\u002Fpricing","jack_pp",[72,37,33,173,11,37],"project\u002Fneutral\u002Fidea",[134,37,44,170,135,33],[157,37,33,170,11,37],[177,37,43,178,135,33],"49115886","project\u002Fpositive\u002Fpricing",[180,37,33,167,11,37],"49116710",[182,183,44,173,184,33],"49120276",2,"h2aichat",[186,183,44,173,187,33],"49125672","kfallah15",[189,183,33,190,11,37],"49127039","project\u002Fneutral\u002Fperformance",[141,44,44,173,142,33],[193,44,33,173,11,37],"49115387",[107,31,37,195,108,33],"project\u002Fcritical\u002Fdocs",[197,31,33,198,11,37],"49063834","project\u002Fneutral\u002Fdocs",[200,43,43,201,202,33],"49063672","project\u002Fpositive\u002Fnone","digitaltrees",[204,43,33,167,11,37],"49063830",[127,32,44,206,128,33],"project\u002Fneutral\u002Fprivacy",[208,32,33,206,11,37],"49063775",[210,211,43,201,212,33],"49063646",7,"rglover",[214,211,33,167,11,37],"49063833",[96,216,37,217,97,33],8,"project\u002Fcritical\u002Fperformance",[100,216,37,219,101,33],"project\u002Fcritical\u002Fidea",[221,216,33,167,11,37],"49063790",[223,216,33,167,224,33],"49063762","irishcoffee",[226,216,33,227,228,33],"49063773","other\u002Fneutral\u002Fnone","teravor",[230,216,33,227,115,33],"49063777",[232,216,33,190,11,37],"49063778",[114,216,37,234,115,33],"project\u002Fcritical\u002Fnone",[236,216,33,237,11,37],"49063822","project\u002Fneutral\u002Fnone",[],[240,244,248,253,258,263],{"id":241,"name":242,"points":243,"category":20},"49098510","TurboFieldfare",919,{"id":245,"name":246,"points":247,"category":20},"49026810","Echo",484,{"id":249,"name":250,"points":251,"category":252},"49021270","blackhole.plav.in",477,"learning_science",{"id":254,"name":255,"points":256,"category":257},"49090607","HNewhere",433,"dev_tools",{"id":259,"name":260,"points":261,"category":262},"49100043","CheapFoodMap",287,"consumer_social",{"id":264,"name":265,"points":266,"category":267},"49057666","Reverse Minesweeper",267,"games",false,"v2",{"id":271,"name":5,"line":6,"ai":268,"cat":20,"year":13,"points":14,"comments":15,"rec":28,"s":272,"top":18,"kw":273,"maker":11,"domain":10,"tier":19,"attempts":33,"strip":274,"url":9,"mix":275,"praise":84,"objection":93,"sub":21},49063454,0.03,[24,25,26,27],"503030503303010503050110000010",[31,32,33,31],1794036706214]