Calibration
τ³-Bench Banking → Harvey LAB
Harvey LAB is estimated from τ³-Bench Banking with a offset logistic curve fitted on 6 models measured on both: y = 0.0089 + (0.9418 − 0.0089) / (1 + exp(−77.26·(x − 0.1178))), R² = 0.97, cross-validated error 2.1 pp. It is used for 4 estimates.
| Estimated model | τ³-Bench Banking | Harvey LAB | Source |
|---|---|---|---|
| K2 Horizon 375B A23B | 34.2% | 94.2% | estimated ± 2.1 pp, medium confidence |
| Qwen3.8 Max | 51.3% | 94.2% | estimated ± 2.1 pp, low confidence |
| Qwen3.8 2.4T A95B | 49.1% | 94.2% | estimated ± 2.1 pp, low confidence |
| Qwen3.8 27B (medium) | 47.4% | 94.2% | estimated ± 2.1 pp, medium confidence |