Calibration
V* → ScreenSpot Pro
ScreenSpot Pro is estimated from V* with a Michaelis–Menten + offset curve fitted on 5 models measured on both: y = 0.0000 + 2.0000·x / (1.93730 + x), R² = 0.62, cross-validated error 7.5 pp. It is used for 1 estimate.
| Estimated model | V* | ScreenSpot Pro | Source |
|---|---|---|---|
| Step 3.7 Flash | 95.3% | 65.9% | estimated ± 7.5 pp, low confidence |
Fitted on
| Model | V* | ScreenSpot Pro |
|---|---|---|
| Claude Opus 4.5 | 67.0% | 45.7% |
| Gemini 3 Pro | 88.0% | 72.7% |
| LFM3-VL-2B | 86.4% | 55.5% |
| Qwen3.5 397B | 95.8% | 65.6% |
| Qwen3.6 Plus | 96.9% | 68.2% |