Calibration
SimpleVQA → ERQA
ERQA is estimated from SimpleVQA with a linear curve fitted on 12 models measured on both: y = 0.5596·x + 0.2699, R² = 0.52, cross-validated error 7.0 pp. It is used for 0 estimates.
Fitted on
| Model | SimpleVQA | ERQA |
|---|---|---|
| Claude Opus 4.5 | 65.7% | 46.8% |
| Gemini 3.1 Pro | 72.4% | 69.4% |
| Gemini 3 Pro | 73.2% | 70.5% |
| GPT-5.2 | 55.8% | 59.8% |
| GPT-5.4 | 61.1% | 65.4% |
| Grok 4.20 | 57.4% | 54.1% |
| LFM3-VL-2B | 36.3% | 48.2% |
| Muse Spark | 71.3% | 64.7% |
| Qwen3.5 397B | 67.1% | 67.5% |
| Qwen3.6-27B | 56.1% | 62.5% |
| Qwen3.7 Plus | 81.7% | 69.8% |
| Qwen3.8 Max | 75.0% | 77.8% |