CogEvol：邁向高效且可靠的學習環境生成

CogEvol: Towards Efficient and Reliable Learning Environment Generation

August 31, 2026

作者: 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

cs.AI

摘要

我們提出 CogEvol，這是一個專為學習環境生成（Learning Environment Generation）所訓練的模型系列：它能將課程概要轉化為完成的學習工件（結構化 JSON 投影片或自包含的互動式 HTML 頁面），且只需單次生成。在 220,000 次生產請求中，CogEvol 生成一張投影片的中位時間為 17 秒，生成一個互動頁面則為 59 秒，取代了耗時數分鐘的多輪代理腳手架。可靠性是透過機制強制保證，而非寄望於偶然：一套基於生產環境的資料管線將真實失敗轉化為 53,687 筆經驗證的 SFT 樣本；結合規則與 VLM 的混合獎勵函數則驅動基於 GRPO 的強化學習，並在我們察覺並修復一次獎勵欺騙事件（該事件產出視覺上具有說服力卻無法遊玩的遊戲）後，此機制獲得進一步加固。CogEvol-27B 在投影片品質評分上取得 83.7 分，在 500 案例的互動式 HTML 基準測試中取得 63.7 分，且參數量僅為旗艦級程式碼模型的 1/26.9；此外，它與 OpenMAIC 團隊合作，為其即時生產流量提供服務。CogEvol-4B 已以 Apache 2.0 授權開源釋出，網址為 https://github.com/CogEvol/CogEvol-4B；外部旗艦模型則在相同測試套件與完全一致的評測框架下進行評測。腳手架編輯進一步將互動式頁面的生成成本降低約 76%；整個技術棧可運行於國產昇騰（Ascend）加速器上，應用層效能與 A800 GPU 相當，從而降低大規模 AI 原生教育的單位成本。

English

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.