CogEvol/CogEvol-4B

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 31, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

CogEvol-4B is a 4.5 billion parameter model developed by CogEvol, specifically post-trained for Learning Environment Generation (LEG). It excels at creating complete learning artifacts, such as structured-JSON slide pages or interactive HTML pages with simulations and exercises, from natural language course briefs. This model is distinguished by its three-stage training recipe on 53,687 verified samples, utilizing a hybrid rule and VLM reward system to ensure measurable interactivity.

Loading preview...

CogEvol-4B: Learning Environment Generation

CogEvol-4B is a 4.5 billion parameter model from the CogEvol family, specialized in Learning Environment Generation (LEG). Its core function is to transform natural language course briefs into fully usable learning artifacts. This includes generating structured-JSON slide pages or self-contained interactive HTML pages that can run directly in a browser, offering simulations, visualizations, and interactive exercises.

Key Capabilities and Features

  • Single-Pass Generation: Creates complete learning environments from a course brief in one go.
  • Diverse Output Formats: Generates both structured-JSON for slides and interactive HTML for dynamic content.
  • Robust Training: Developed using a three-stage recipe (mix SFT → Slide RL → interactive-HTML RL) on 53,687 verified samples.
  • Reliable Interactivity: Employs a hybrid rule-based and Vision-Language Model (VLM) reward system, hardened against reward hacking, where interactivity is measured by automated probes rather than subjective judgment.
  • Efficient Deployment: Optimized for local deployment, with a GitHub repository providing a complete llama.cpp setup, validated serving flags, and integration with the OpenMAIC app for offline use. A quantized Q4_K_M GGUF version (~2.4 GB) is also available.

Use Cases

CogEvol-4B is ideal for developers and educators looking to automate the creation of interactive and structured learning content. It is particularly suited for generating:

  • Interactive educational modules.
  • Automated slide presentations for courses.
  • Browser-based simulations and visualizations for learning.
  • Self-contained exercises for educational platforms.

Finetunes
3 models