benchgap
Calibration

ScreenSpot Pro → CharXiv

CharXiv is estimated from ScreenSpot Pro with a Michaelis–Menten curve fitted on 12 models measured on both: y = 1.2249·x / (0.35656 + x), R² = 0.71, cross-validated error 3.9 pp. It is used for 6 estimates.

Estimated modelScreenSpot ProCharXivSource
Claude Opus 4.683.1%85.7%estimated ± 3.9 pp, high confidence
GPT-6 Astra92.7%88.5%estimated ± 3.9 pp, medium confidence
Holo2-235B-A22B70.6%81.4%estimated ± 3.9 pp, high confidence
Holo2-30B-A3B66.1%79.6%estimated ± 3.9 pp, high confidence
Holo2-4B57.2%75.5%estimated ± 3.9 pp, high confidence
Holo2-8B58.9%76.3%estimated ± 3.9 pp, high confidence