Claim-level Confidence Calibration on BeyondAIME
0.301SNR Gain
Qwen3-4B-Instruct-confidence-min
Updated 4d ago
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| 0.301 | 78.8 | 61.6 | 25.8 | 25.1 | 0.81 | 53.7 | ||
| 0.183 | 74.1 | 52.6 | 31.1 | 30.5 | 0.93 | 49.7 | ||
2025.12 | 0.068 | 60.1 | 77.4 | 21 | 21.1 | 0.887 | 77 | |
2025.12 | 0.029 | 67.6 | 71.9 | 29.4 | 26 | 1.027 | 71.9 | |
2025.12 | 0.027 | 61.7 | 56.3 | 45.6 | 41.9 | 1.809 | 55.8 | |
2025.12 | 0.019 | 71.9 | 89.1 | 13 | 10.4 | 0.449 | 89.1 | |
2025.12 | 0.018 | 71.7 | 88.9 | 13.2 | 10.6 | 0.469 | 88.9 |