[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-vision-encoder-shrinks-video-ai-tokens-without-losing-detail":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},9434,"new-vision-encoder-shrinks-video-ai-tokens-without-losing-detail","New Vision Encoder Shrinks Video AI Tokens Without Losing Detail","A new codec-native encoder compresses long videos into as few as 400 visual tokens, cutting AI processing costs without sacrificing understanding.","Researchers have built a vision encoder that processes an entire video in one pass and still boils it down to a few hundred tokens.\n\nThe system, called CoVisco, reads video directly in its native codec rather than decoding every frame into dense patches first. It splits footage into temporal segments, each with its own set of learnable \"abstract tokens\" that summarize that chunk, while the original patch-level detail stays accessible if needed. A selector then decides whether a downstream model gets just the abstract tokens or those plus a handful of relevant patches. The team trained it on 565 million image-text pairs and 6.4 million videos, and in a test using 64 frames split into four segments, the abstract-only version needed just 400 visual tokens to match or beat OneVision-Encoder on several video benchmarks.\n\nThat token count matters more than it sounds. Every visual token a language model has to read eats into context length and slows down the \"prefill\" step before it can answer anything, which is why long-video understanding has stayed expensive even as models get better. Most fixes compress tokens after the fact, bolted onto a model trained on dense representations - CoVisco instead builds the compact interface into training itself, so there's no mismatch between what the model learned and what it has to work with at inference time.\n\nWhether 400 tokens is actually enough context for gnarly video-reasoning tasks, as opposed to the benchmarks chosen for the paper, is the kind of question that only gets answered once other labs start poking at the code on GitHub.","[\"ai\",\"computer-vision\",\"video-understanding\",\"machine-learning\"]","2026-10-01T04:00:00.000Z","2026-10-02T18:38:24.715Z","2026-10-02T18:38:29.492Z","published",null,[],"ai",[24,26,27,28],"computer-vision","video-understanding","machine-learning",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.39924",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,90,95,100],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",5748,"2026-10-02T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",829,"2026-10-01T17:31:56.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",437,"2026-10-01T18:10:00.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",317,"2026-10-01T22:00:00.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",198,"2026-10-01T17:38:48.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Science","science",168,"2026-10-01T18:35:55.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",155,"2026-10-01T19:54:10.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Dev Tools","dev-tools",96,"2026-10-01T16:57:03.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Software","software",93,"2026-09-30T21:41:11.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Startups","startups",90,"2026-10-01T21:55:22.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",53,"2026-10-02T02:50:39.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]