[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-test-on-device-translation-for-live-chat":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},9222,"researchers-test-on-device-translation-for-live-chat","Researchers Test On-Device Translation for Live Chat","Researchers built a Korean-English benchmark showing on-device models can match GPT-5.1 on live-stream chat translation, if phones can handle the load.","A new study finds phone-sized AI models can translate live-stream chat about as well as GPT-5.1, provided you pick the right model for the right phone.\n\nResearchers built LiveChatBench, a benchmark of 1,000 Korean-English sentence pairs drawn from live-stream chat messages, to test on-device translation models under real deployment conditions. They ran the models on five actual mobile devices, tracking not just translation accuracy but CPU load and thermal behavior, the kind of operational detail that rarely shows up in model benchmarks. After adapting models to the live-chat domain, the best-performing on-device setups reached translation quality comparable to GPT-5.1 on this specific task. Model selection mattered a lot: different phones and model sizes produced very different resource tradeoffs, meaning there is no single on-device model that works everywhere.\n\nLive chat translation is exactly the kind of workload where cloud APIs are a poor fit: it has to run constantly, in real time, for potentially millions of simultaneous viewers, without racking up per-message API costs or sending chat logs to a third-party server. This study is useful precisely because it treats phone heat and processor strain as real constraints instead of footnotes, something most on-device AI papers skip. It suggests narrow, well-scoped translation tasks are where small local models can already compete with frontier cloud models, even if they cannot match them on general-purpose translation.\n\nThat caveat matters: a model tuned to beat GPT-5.1 on 1,000 Korean-English chat lines is not the same as a model that translates anything well, and the paper itself frames the result as task-specific rather than evidence that on-device AI has broadly caught up to cloud models.","[\"on-device-ai\",\"translation\",\"live-streaming\",\"mobile-devices\"]","2026-10-01T04:00:00.000Z","2026-10-02T02:40:10.241Z","2026-10-02T02:40:11.527Z","published",null,[],"ai",[26,27,28,29],"on-device-ai","translation","live-streaming","mobile-devices",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2601.02641",0,{"sections":36},[37,40,44,48,53,58,62,67,72,76,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5629,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",816,{"name":45,"slug":46,"count":47,"latest_published_at":18},"Policy","policy",430,{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",163,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":73,"slug":74,"count":70,"latest_published_at":75},"Software","software","2026-09-30T21:41:11.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]