AI summaryⓘ
The authors introduce CogEvol, a set of models designed to quickly turn course outlines into finished learning materials like slides or interactive web pages in one step. They tested it on many requests, showing it works reliably and much faster than previous methods that needed multiple interactions. The team improved reliability using a special data pipeline and a mix of rules and learning-based rewards, fixing issues that made some outputs look good but not work. Their largest model performs well on quality tests with fewer parameters than big coding models, and a smaller version is publicly available. They also reduced costs using efficient editing methods and hardware that matches more expensive GPUs, aiming to make AI-driven education cheaper and more practical.
Learning Environment Generationstructured JSON slidesinteractive HTML pagesreinforcement learning (RL)reward hackingGRPO (reward optimization)supervised fine-tuning (SFT)parameter efficiencyAscend acceleratorsApache 2.0 license
Authors
Shangqing Tu, Daniel Zhang-Li, Yucheng Wang, Shiyu Gan, Yanpeng Wang, Huiqiang Rong, Mofei Chen, Shen Yang, Yini Chen, Yinuo Duan, Haoxuan Li, Binglin Liu, Ye He, Danqi Zheng, Zhanxin Hao, Yuxuan Wu, Mengting Tao, Yuqiu Liu, Jifan Yu, Juanzi Li, Bin Xu, Lei Hou, Huiqin Liu, Yu Zhang
Abstract
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.