MalruleLib - Cross-Template with Steps: leaderboard
Cross-template misconception prediction from MalruleLib (Chen, Liu and Sonkar, ACL 2026) when the worked example also shows the student's step-by-step malrule reasoning: the model predicts the answer the same student gives on a problem from a different template of that malrule. Exact-match accuracy in percent over 7,700 pairs; higher is better.
Metric: Malrule Reasoning Accuracy (%; predict the student's next answer from one worked mistake; cross-template, with the student's steps). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-OSS-120B | 60.9 | #699 |
| 2 | Qwen 3 Next 80B A3B (Thinking) | 59.8 | #591 |
| 3 | GPT-OSS-20B | 57.7 | #1150 |
| 4 | Qwen 3 Next 80B A3B Instruct | 54.4 | #763 |
| 5 | Llama 3.3 70B Instruct | 48.5 | #1353 |
| 6 | Phi-4 | 47.2 | #1920 |
| 7 | Qwen 3 4B | 34.6 | #2297 |
| 8 | Phi-4 Mini Instruct | 26.3 | #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-cross-template-with-steps · How It Works · Data refreshed daily, snapshot 2026-11-12.