CogEvol introduces a family of models designed to convert course briefs into structured JSON slides or interactive HTML pages in a single pass, replacing multi-turn agent scaffolding. The system utilizes a production-grounded data pipeline and hybrid reward mechanisms to enforce reliability during training.

  • CogEvol completes slide generation with a median latency of 17 seconds and interactive page generation in 59 seconds across 220k requests.
  • CogEvol-27B achieves scores of 83.7 on slide quality and 63.7 on a 500-case interactive HTML benchmark using 26.9x fewer parameters than flagship coding models.
  • CogEvol-4B is released openly under the Apache 2.0 license, while CogEvol-27B serves live production traffic in collaboration with the OpenMAIC team.
  • Scaffold editing reduces interactive-page generation costs by approximately 76%, and the stack runs on domestic Ascend accelerators at parity with A800 GPUs.

This approach lowers the unit cost of AI-native education at scale by enabling efficient, reliable artifact generation on accessible hardware.