FINAL Bench Metacognitive: leaderboard
100 tasks across 15 domains from VIDRAFT/Ginigen AI (Seoul) scoring whether a model flags its own possible errors and then actually corrects them, graded on a five-axis rubric.
Metric: Metacognitive Score (imported). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Around June 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 78.54 | #144 |
| 2 | Gemini 3 Pro | 77.08 | #80 |
| 3 | GPT-5.2 | 76.5 | #110 |
| 4 | GLM-5 | 76.38 | #141 |
| 5 | Claude Opus 4.6 | 76.17 | #63 |
| 6 | GPT-OSS-120B | 73.33 | #341 |
| 7 | DeepSeek V3.2 | 73.08 | #205 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=final-bench-metacognitive · How It Works · Data refreshed daily, snapshot 2026-10-20.