Search papers, labs, and topics across Lattice.
CogEvol is a suite of models designed for efficient Learning Environment Generation, transforming course briefs into structured learning artifacts in a single pass. By leveraging a production-grounded data pipeline and a hybrid rule-plus-VLM reward system, CogEvol achieves a median slide completion time of 17 seconds and an interactive page time of 59 seconds, significantly reducing the time and cost compared to traditional methods. The models demonstrate high reliability and quality, scoring 83.7 on slide quality and 63.7 on interactive-HTML benchmarks, while operating with 26.9x fewer parameters than leading coding models.
CogEvol slashes learning artifact generation time from minutes to seconds while ensuring high reliability and quality.
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.
