Calibration
VITA-Bench → BrowseComp
BrowseComp is estimated from VITA-Bench with a linear curve fitted on 6 models measured on both: y = 0.9383·x + 0.3350, R² = 0.81, cross-validated error 8.0 pp. It is used for 0 estimates.
Fitted on
| Model | VITA-Bench | BrowseComp |
|---|---|---|
| Agents-A1 | 38.8% | 75.5% |
| Agents-A1-4B | 40.3% | 66.8% |
| GLM-4.7 | 15.5% | 52.0% |
| LongCat-Flash-Lite-Sparse | 21.7% | 48.6% |
| Qwen3.5 397B | 43.7% | 69.0% |
| Sage-A1.5 | 45.4% | 81.7% |