Calibration
CharXiv → ScreenSpot Pro
ScreenSpot Pro is estimated from CharXiv with a Hill curve fitted on 12 models measured on both: y = 0.0000 + (1.0701 − 0.0000)·x^6.00 / (0.71035^6.00 + x^6.00), R² = 0.75, cross-validated error 7.1 pp. It is used for 15 estimates.
| Estimated model | CharXiv | ScreenSpot Pro | Source |
|---|---|---|---|
| Claude Mythos 5 | 93.5% | 89.8% | estimated ± 7.1 pp, medium confidence |
| Claude Opus 4.7 (Adaptive) | 91.0% | 87.3% | estimated ± 7.1 pp, medium confidence |
| Claude Sonnet 4.6 | 77.4% | 67.0% | estimated ± 7.1 pp, medium confidence |
| Claude Sonnet 5 | 88.3% | 84.2% | estimated ± 7.1 pp, medium confidence |
| Command A+ | 52.7% | 15.3% | estimated ± 7.1 pp, low confidence |
| Gemini 3.1 Flash-Lite | 73.2% | 58.3% | estimated ± 7.1 pp, medium confidence |
| Gemini 3.5 Flash | 84.2% | 78.7% | estimated ± 7.1 pp, medium confidence |
| Gemini 3.7 Flash | 88.7% | 84.7% | estimated ± 7.1 pp, medium confidence |
| GLM-5.3-Flash | 89.4% | 85.5% | estimated ± 7.1 pp, medium confidence |
| Inkling | 82.0% | 75.2% | estimated ± 7.1 pp, medium confidence |
| Inkling-Small | 81.3% | 74.1% | estimated ± 7.1 pp, medium confidence |
| MiMo-V2.5 | 81.0% | 73.6% | estimated ± 7.1 pp, medium confidence |
| Muse Spark 1.1 | 88.4% | 84.3% | estimated ± 7.1 pp, medium confidence |
| Sakana Fugu | 85.1% | 80.0% | estimated ± 7.1 pp, medium confidence |
| Sakana Fugu-Ultra | 86.6% | 82.0% | estimated ± 7.1 pp, medium confidence |