CogEvol-4B turns a one-line brief into a complete interactive lesson: structured slides plus self-contained HTML simulations, in a single forward pass, running fully offline on a laptop with no APIs, no cloud, and no agent loops😁<br><a href=\"https://cdn-uploads.huggingface.co/production/uploads/648c48d8c0ddeee6df5b6d22/03oo4hrKxCzrkqpb2vl6t.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/648c48d8c0ddeee6df5b6d22/03oo4hrKxCzrkqpb2vl6t.png\" alt=\"showcase_prod_en_preview\"></a></p>\n","updatedAt":"2026-09-01T04:05:55.151Z","author":{"_id":"648c48d8c0ddeee6df5b6d22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c48d8c0ddeee6df5b6d22/BlrYDv3eQxZ-Y5vtVGegX.jpeg","fullname":"Shangqing Tu","name":"tsq2000","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":6,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6877402067184448},"editors":["tsq2000"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/648c48d8c0ddeee6df5b6d22/BlrYDv3eQxZ-Y5vtVGegX.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.30968","authors":[{"_id":"6a963c64cd6ebc484732ec35","name":"Shangqing Tu","hidden":false},{"_id":"6a963c64cd6ebc484732ec36","name":"Daniel Zhang-Li","hidden":false},{"_id":"6a963c64cd6ebc484732ec37","name":"Yucheng Wang","hidden":false},{"_id":"6a963c64cd6ebc484732ec38","name":"Shiyu Gan","hidden":false},{"_id":"6a963c64cd6ebc484732ec39","name":"Yanpeng Wang","hidden":false},{"_id":"6a963c64cd6ebc484732ec3a","name":"Huiqiang Rong","hidden":false},{"_id":"6a963c64cd6ebc484732ec3b","name":"Mofei Chen","hidden":false},{"_id":"6a963c64cd6ebc484732ec3c","name":"Shen Yang","hidden":false},{"_id":"6a963c64cd6ebc484732ec3d","name":"Yini Chen","hidden":false},{"_id":"6a963c64cd6ebc484732ec3e","name":"Yinuo Duan","hidden":false},{"_id":"6a963c64cd6ebc484732ec3f","name":"Haoxuan Li","hidden":false},{"_id":"6a963c64cd6ebc484732ec40","name":"Binglin Liu","hidden":false},{"_id":"6a963c64cd6ebc484732ec41","name":"Ye He","hidden":false},{"_id":"6a963c64cd6ebc484732ec42","name":"Danqi Zheng","hidden":false},{"_id":"6a963c64cd6ebc484732ec43","name":"Zhanxin Hao","hidden":false},{"_id":"6a963c64cd6ebc484732ec44","name":"Yuxuan Wu","hidden":false},{"_id":"6a963c64cd6ebc484732ec45","name":"Mengting Tao","hidden":false},{"_id":"6a963c64cd6ebc484732ec46","name":"Yuqiu Liu","hidden":false},{"_id":"6a963c64cd6ebc484732ec47","name":"Jifan Yu","hidden":false},{"_id":"6a963c64cd6ebc484732ec48","name":"Juanzi Li","hidden":false},{"_id":"6a963c64cd6ebc484732ec49","name":"Bin Xu","hidden":false},{"_id":"6a963c64cd6ebc484732ec4a","name":"Lei Hou","hidden":false},{"_id":"6a963c64cd6ebc484732ec4b","name":"Huiqin Liu","hidden":false},{"_id":"6a963c64cd6ebc484732ec4c","name":"Yu Zhang","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/648c48d8c0ddeee6df5b6d22/ZSpN_apoOfdJi_fBAwOYH.mp4"],"publishedAt":"2026-08-31T00:00:00.000Z","submittedOnDailyAt":"2026-09-01T00:00:00.000Z","title":"CogEvol: Towards Efficient and Reliable Learning Environment Generation","submittedOnDailyBy":{"_id":"648c48d8c0ddeee6df5b6d22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c48d8c0ddeee6df5b6d22/BlrYDv3eQxZ-Y5vtVGegX.jpeg","isPro":false,"fullname":"Shangqing Tu","user":"tsq2000","type":"user","name":"tsq2000"},"summary":"We present CogEvol, a family of models trained specifically for Learning Environment Generation: turning a course brief into a finished learning artifact (structured-JSON slides or self-contained interactive HTML pages) in a single pass. Across 220k production requests, CogEvol completes a slide in a median of 17 seconds and an interactive page in 59, replacing minutes-long multi-turn agent scaffolding. Reliability is enforced rather than hoped for: a production-grounded data pipeline turns real failures into 53,687 verified SFT samples, and a hybrid rule-plus-VLM reward drives GRPO-based RL, hardened after we caught and fixed a reward-hacking episode that produced visually convincing but unplayable games. CogEvol-27B scores 83.7 on slide quality and 63.7 on a 500-case interactive-HTML benchmark with 26.9x fewer parameters than flagship coding models, and, in collaboration with the OpenMAIC team, serves their live production traffic. CogEvol-4B is released openly under the Apache 2.0 license at https://github.com/CogEvol/CogEvol-4B; external flagships are measured on the same suites under the identical harness. Scaffold editing cuts interactive-page generation cost by a further ~76%, and the full stack runs on domestic Ascend accelerators at application-level parity with A800 GPUs, lowering the unit cost of AI-native education at scale.","upvotes":21,"discussionId":"6a963c65cd6ebc484732ec4d","githubRepo":"https://github.com/CogEvol/CogEvol-4B","githubRepoAddedBy":"user","ai_summary":"CogEvol is a family of models that generate structured learning artifacts in a single pass using supervised fine-tuning and reinforcement learning with vision-language rewards, achieving high quality with far fewer parameters and lower cost.","ai_keywords":["SFT","GRPO","reinforcement learning","VLM reward","reward hacking"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":1,"organization":{"_id":"6a34d943d09c02d62f86b2bb","name":"CogEvol","fullname":"元知进化（北京）科技有限公司","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c48d8c0ddeee6df5b6d22/z95C6e3gbYmdwNxinzUbC.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"648c48d8c0ddeee6df5b6d22","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c48d8c0ddeee6df5b6d22/BlrYDv3eQxZ-Y5vtVGegX.jpeg","isPro":false,"fullname":"Shangqing Tu","user":"tsq2000","type":"user"},{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"648c5c1d4119ec0d467ef612","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c5c1d4119ec0d467ef612/Hn1JN6ETrijlcsMdOzO4n.png","isPro":false,"fullname":"Daniel Zhang-Li","user":"JagdishDZ","type":"user"},{"_id":"632d6e8d2f9c892dcf0ed4bb","avatarUrl":"/avatars/fba1fb87bcac340a8eee3b9f4fc35bc5.svg","isPro":false,"fullname":"Jifan Yu","user":"JovanYu","type":"user"},{"_id":"66a894e46419b5b8fc03a2d6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/WaZJTfPS07Ck88A7CFIhn.png","isPro":false,"fullname":"lijinhe","user":"Lijinhe","type":"user"},{"_id":"66ed273bc890e8ac7153020c","avatarUrl":"/avatars/f1ee45fae4e57776fbc8e1a2255776e0.svg","isPro":false,"fullname":"Huiqiang Rong","user":"Eric1231","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"659bb25ce6df49a09d637be4","avatarUrl":"/avatars/da12bd2028c2113116c862b768d3a654.svg","isPro":false,"fullname":"Lucas Li","user":"LucasLi666","type":"user"},{"_id":"6718e4b1a8065b01d0221778","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6718e4b1a8065b01d0221778/32k6U-rNbsgtV8NqwghVN.jpeg","isPro":false,"fullname":"严舞雩","user":"freshlittlelemon","type":"user"},{"_id":"66cdd285c51a915bd5f2d017","avatarUrl":"/avatars/14e5794307e4672b1b51d26b31227e0f.svg","isPro":false,"fullname":"Jiajie Zhang","user":"NeoZ123","type":"user"},{"_id":"67f1e3085542f471d9598114","avatarUrl":"/avatars/37db1c8123a0f4ac30e06713fb9eed27.svg","isPro":false,"fullname":"Leo","user":"anticyc","type":"user"},{"_id":"68be24090a3fcebdcad43ec3","avatarUrl":"/avatars/6d35cee159dbe991258e640b406b9ad4.svg","isPro":false,"fullname":"Jie Cao","user":"Xiaojiecao","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6a34d943d09c02d62f86b2bb","name":"CogEvol","fullname":"元知进化（北京）科技有限公司","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/648c48d8c0ddeee6df5b6d22/z95C6e3gbYmdwNxinzUbC.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.30968.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2608.30968 in a dataset README.md to link it from this page.

Add this paper to a collection to link it from this page.