MalruleLib - Forward MRA: leaderboard
Forward misconception application from MalruleLib (Chen, Liu and Sonkar, ACL 2026): given a written description of a documented student misconception and a new problem, the model predicts the wrong answer a student holding that misconception would give. Exact-match accuracy in percent over 4,991 problems covering 101 misconceptions; higher is better.
Metric: Forward MRA Accuracy (%; predict the student's answer from a description of the misconception). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 Next 80B A3B (Thinking) | 53.9 | #591 |
| 2 | GPT-OSS-120B | 48.8 | #699 |
| 3 | GPT-OSS-20B | 46.5 | #1150 |
| 4 | Llama 3.3 70B Instruct | 39.6 | #1353 |
| 5 | Phi-4 | 37.4 | #1920 |
| 6 | Qwen 3 Next 80B A3B Instruct | 30.6 | #763 |
| 7 | Qwen 3 4B | 17.4 | #2297 |
| 8 | Phi-4 Mini Instruct | 9.3 | #6452 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=malrulelib-forward-mra · How It Works · Data refreshed daily, snapshot 2026-11-12.