Positioning spread
| Benchmark | Task acc | Confidence | F₁ | Leans |
|---|---|---|---|---|
| SQuAD (factual recall) | 0.65 | 0.68 | 0.79 | +0.03 calibrated |
| MMLU-Pro (knowledge) | 0.66 | 0.79 | 0.81 | +0.13 overconfident |
| LegalBench (legal reasoning) | 0.87 | 0.96 | 0.93 | +0.10 overconfident |
| MathBench (competition math) | 1.00 | 1.00 | 1.00 | +0.00 calibrated |
| OmniMath (advanced math) | 0.88 | 0.98 | 0.94 | +0.10 overconfident |
| SciCode (scientific code) | 0.64 | 0.48 | 0.62 | -0.15 cautious |
In the full cloud
Gemini 3 Pro conditions all other model/condition points equal relative confidence and pass rate
Pairwise footprint
Match accuracy controls for the performance base-rate gap
Gemini 3 Pro pairs
18/ 171
Gemini 3 Pro mean tau
+0.086
All-pairs mean
+0.037
Gemini 3 Pro p<0.05
16(88.9%)
all model pairs (observed) base-rate-matched null calibration-preserving null Gemini 3 Pro pair (filled = p<0.05) Gemini 3 Pro mean all-pairs mean