MMCL-Bench - Rubric Accuracy: leaderboard
Metric: Rubric-level accuracy (%): share of all rubric items satisfied across tasks, on the 102 MMCL-Bench tasks (49 rule-system, 32 procedural, 21 empirical), in which a model learns a task-local rule, procedure or pattern from visual or mixed-modality teaching context and applies it to a new visual instance, scored against expert rubrics by a rubric-only LLM judge, models at their recommended or default decoding settings; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Thinking) | 76.6 |
| 2 | Gemini 3.1 Pro (Preview) | 67.6 |
| 3 | Qwen 3.6 Plus (Thinking) | 67.6 |
| 4 | Claude Opus 4.6 (Thinking) | 66 |
| 5 | Kimi K2.5 (Thinking) | 64.1 |
| 6 | GPT-5.4 | 59.9 |
Interactive version: theaggregate.ai/benchmark?slug=mmcl-bench-rubric-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-07.