A benchmark suite for evaluating Executive Function capabilities of AI models using Wordle game and Information Gain optimization
Unlike traditional "plain-vanilla" win-rate benchmarks, this framework measures the working memory drift, cognitive efficiency, and constraint-adherence of AI agents under dynamic restrictions.
It is based on the kaggle benchmarks and allows cross-task insights showing the potential of wordle game as an efficient benchmark for Executive Function of AI Models
- Working Memory: Context retention and rule-drift resistance over multi-turn interactions.
- Cognitive Flexibility: Adaptive reasoning when shifting strategies dynamically.
- Inhibitory Control: Impulse suppression and mistake minimization under stress.
Explore the full, interactive cross-task leaderboard, adjust scoring weights dynamically, and view normalized model cognitive maps here: 👉 Live Dashboard
This dashboard give more metrics and insights based on kaggle benchmark task outputs.
/data- Standardized JSON evaluation outputs for single-turn, multi-turn, and cognitive flexibility tasks./templates- HTML UI layout templates.generate.py- Core Python pipeline for processing Kaggle leaderboard metrics and generating the static site.
- Data Source: All metrics are extracted from standardized benchmark notebook runs on the Kaggle Benchmarks Platform