Calibration
SimpleVQA → ZeroBench
ZeroBench is estimated from SimpleVQA with a linear curve fitted on 5 models measured on both: y = -1.3013·x + 1.2089, R² = 0.69, cross-validated error 6.4 pp. It is used for 5 estimates.
| Estimated model | SimpleVQA | ZeroBench | Source |
|---|---|---|---|
| Grok 4.20 | 57.4% | 46.2% | estimated ± 6.4 pp, low confidence |
| LFM2.5-VL-3B | 35.4% | 74.8% | estimated ± 6.4 pp, low confidence |
| Qwen3.6-27B | 56.1% | 47.9% | estimated ± 6.4 pp, low confidence |
| Qwen3.6-35B-A3B | 58.9% | 44.2% | estimated ± 6.4 pp, low confidence |
| Step 3.7 Flash | 79.2% | 17.8% | estimated ± 6.4 pp, low confidence |