AdaExploreBench (Evidence Given) - Calibrated Accuracy: leaderboard
AdaExploreBench's calibrated accuracy when the evaluation batches are unmasked, a diagnostic setting: evidence about the hidden rule accumulates passively instead of having to be sought with probes, so the gap to the main setting measures the cost of active exploration.
Metric: CA (%) (0-100). Source: hkust-knowcomp.github.io. Saturation forecast: Around February 2027. 6 models tracked. Updated 19 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | High | 70.34 | #69 |
| 2 | GPT-6 Sol | High | 66.57 | #33 |
| 3 | GPT-6 Luna | High | 51.53 | #106 |
| 4 | Gemma 4 12B (IT) | High | 36.52 | #731 |
| 5 | Qwen 3.8 27B | Medium | 33.78 | #220 |
| 6 | Qwen 3.5 9B | 28.18 | #829 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=adaexplorebench-evidence-given-calibrated-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-20.