Calibration
CyberGym → BrowseComp
BrowseComp is estimated from CyberGym with a Michaelis–Menten curve fitted on 13 models measured on both: y = 2.0000·x / (1.09826 + x), R² = 0.49, cross-validated error 5.2 pp. It is used for 6 estimates.
| Estimated model | CyberGym | BrowseComp | Source |
|---|---|---|---|
| Gemini 3.5 Flash Cyber | 83.2% | 86.2% | estimated ± 5.2 pp, low confidence |
| Gemini 3.8 Flash Cyber | 86.2% | 87.9% | estimated ± 5.2 pp, low confidence |
| Hy4 preview | 78.4% | 83.3% | estimated ± 5.2 pp, low confidence |
| Ling 3.1 Flash | 87.9% | 88.9% | estimated ± 5.2 pp, low confidence |
| Muse Spark | 43.5% | 56.7% | estimated ± 5.2 pp, low confidence |
| Fugu Cyber | 86.9% | 88.3% | estimated ± 5.2 pp, low confidence |