Metacognitive Monitoring Battery: leaderboard
Six developmental-psychology paradigms over 524 items (overhypothesis induction, signal-detection metacognitive calibration, mutual exclusivity, biased competition, magnitude and executive function, prospective regulation), scored as the mean of the six task scores. Tests whether a model tracks its own accuracy well enough to withdraw a wrong answer. Higher is better. Source: arxiv.org/abs/2604.15702.
Metric: Mean Task Score (%). Source: www.kaggle.com. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 94.33 | #90 |
| 2 | Claude Opus 4.6 | 93.76 | #67 |
| 3 | GLM-5 | 93.62 | #149 |
| 4 | Gemini 3.1 Pro (Preview) | 93.33 | #64 |
| 5 | Gemini 2.5 Flash | 93.23 | #255 |
| 6 | Claude Sonnet 4.6 | 92.73 | #94 |
| 7 | Gemini 2.5 Pro | 92.03 | #159 |
| 8 | GPT-5.4 | 91.84 | #83 |
| 9 | Qwen 3 235B A22B 2507 Instruct | 89.78 | #315 |
| 10 | Claude Haiku 4.5 | 88.29 | #287 |
| 11 | GPT-5.4 Mini | 87.67 | #228 |
| 12 | Qwen 3 Coder 480B A35B Instruct | 87.22 | #314 |
| 13 | DeepSeek V3.2 | 84.66 | #210 |
| 14 | Gemma 3 27B | 83.27 | #606 |
| 15 | Gemma 3 12B | 83.1 | #674 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=metacognitive-monitoring-battery · How It Works · Data refreshed daily, snapshot 2026-11-05.