Calibration
SimpleVQA → Video-MME (with subtitle)
Video-MME (with subtitle) is estimated from SimpleVQA with a Michaelis–Menten curve fitted on 10 models measured on both: y = 1.1257·x / (0.20234 + x), R² = 0.65, cross-validated error 5.7 pp. It is used for 0 estimates.
Fitted on
| Model | SimpleVQA | Video-MME (with subtitle) |
|---|---|---|
| Claude Opus 4.5 | 65.7% | 77.6% |
| Gemini 3 Pro | 73.2% | 88.4% |
| GPT-5.2 | 55.8% | 86.0% |
| Kimi K2.5 | 71.2% | 87.4% |
| LFM3-VL-2B | 36.3% | 69.0% |
| Qwen3.5 397B | 67.1% | 87.5% |
| Qwen3.6-27B | 56.1% | 87.7% |
| Qwen3.6-35B-A3B | 58.9% | 86.6% |
| Qwen3.7 Plus | 81.7% | 88.0% |
| Qwen3.8 Max | 75.0% | 90.4% |