Calibration on Math datasets Aggregated models
0.058ECE2Supervised Platt scaling (Sup.)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Supervised Platt scaling (Sup.)2026.04 | 0.058 | 0.11 | |
| Unsupervised Confidence CalibrationEmbeddings=Internal2026.04 | 0.075 | 0.102 | |
| Unsupervised Confidence CalibrationEmbeddings=External (Black-Box)2026.04 | 0.077 | 0.1 | |
| Test-time self-consistency (TT SC)2026.04 | 0.078 | 0.083 | |
| Token Probs. (TP)2026.04 | 0.116 | 0.128 | |
| Verbalized Confidence (VC)2026.04 | 0.165 | 0.15 | |
| Answer Probs. (Ans. TP)2026.04 | 0.182 | 0.147 |