[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-small-ai-coding-agents-match-models-136-times-their-size":10,"sections":45},{"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":34,"tags":35,"sources":40,"feedback":44,"feedback_at":22,"cost_usd":44,"total_tokens":44},9335,"small-ai-coding-agents-match-models-136-times-their-size","Small AI Coding Agents Match Models 13.6 Times Their Size","A new training method teaches small coding agents to use files as memory, letting a 9B model match one 13.6 times its size on coding benchmarks.","A 9-billion-parameter AI coding agent is now holding its own against a model 13.6 times its size, thanks to a new way of training it to take notes.\n\nResearchers built a training suite called Coding Agent Memory Gym (CAMG), covering four long-horizon task types: shopping, coding, deep research, and autonomous research. Instead of giving the agent a custom memory tool, they gave it shell access and a persistent workspace, letting it create, edit, and search its own files as memory. They then trained a single policy across all four environments with reinforcement learning, using a method called asynchronous PPO, rewarding the agent only for finishing tasks - not for managing memory any particular way. The resulting models, CAMG-RL-4B and CAMG-RL-9B, were built from matching-size Qwen3.5 base models.\n\nOn two established benchmarks, SWE-bench Verified and MLE-bench Lite, the 9B version performed about as well as Qwen3.5-122B-A10B, a model 13.6 times larger. The 4B version matched Qwen3.5-35B-A3B, a model nearly nine times its size. That gap matters because smaller models are cheaper to run and easier to deploy without a server farm backing them up.\n\nBenchmark parity is not real-world parity, and every claim that a small model beats a big one eventually meets a less forgiving task outside the test set.","[\"ai agents\",\"reinforcement learning\",\"coding agents\",\"benchmarks\"]","2026-10-01T04:00:00.000Z","2026-10-02T10:07:38.590Z","2026-10-02T10:07:39.702Z","published",null,[24,30],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Fix the dek's 'nearly ten times' claim: the 4B model is compared to Qwen3.5-35B-A3B, which is 8.75x its size (not ten), while the 9B-to-122B pairing is about 13.6x, so use an accurate multiplier for each comparison instead of one vague rounded figure.","resolved",{"id":31,"reviewer":26,"round":32,"reason":33,"status":29},"editor-r2",2,"Fix the headline's 'Up to 13 Times Their Size' — the only multiplier in the piece is 13.6x, so 'up to 13' understates it; round to 'nearly 14 times' or state '13.6 times' to match the dek and body.","ai",[36,37,38,39],"ai agents","reinforcement learning","coding agents","benchmarks",[41],{"name":42,"url":43},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.34422",0,{"sections":46},[47,51,56,61,66,71,75,80,85,89,93,98,103,108],{"name":48,"slug":34,"count":49,"latest_published_at":50},"AI",5694,"2026-10-01T12:05:27.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Security","security",821,"2026-10-01T14:00:00.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Policy","policy",431,"2026-10-01T11:08:42.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Deals","deals",311,"2026-10-01T14:19:00.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Hardware","hardware",197,"2026-10-01T11:37:06.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":18},"Science","science",165,{"name":76,"slug":77,"count":78,"latest_published_at":79},"Consumer Tech","consumer-tech",150,"2026-10-01T11:59:27.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Software","software","2026-09-30T21:41:11.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":55},"Startups","startups",86,{"name":94,"slug":95,"count":96,"latest_published_at":97},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":104,"slug":105,"count":106,"latest_published_at":107},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":109,"slug":110,"count":111,"latest_published_at":112},"How-To","how-to",7,"2026-10-01T09:00:00.000Z"]