Calibration
VideoMMMU → MLVU (M-Avg)
MLVU (M-Avg) is estimated from VideoMMMU with a linear curve fitted on 10 models measured on both: y = 0.4058·x + 0.5111, R² = 0.80, cross-validated error 2.8 pp. It is used for 0 estimates.
Fitted on
| Model | VideoMMMU | MLVU (M-Avg) |
|---|---|---|
| Claude Opus 4.5 | 84.4% | 81.7% |
| Gemini 3 Pro | 87.6% | 83.0% |
| GPT-5.2 | 85.9% | 85.6% |
| Kimi K2.5 | 86.6% | 85.0% |
| LFM3-VL-2B | 49.6% | 71.3% |
| Qwen3.5 397B | 84.7% | 86.7% |
| Qwen3.6-27B | 84.4% | 86.6% |
| Qwen3.6-35B-A3B | 83.7% | 86.2% |
| Qwen3.7 Plus | 85.4% | 87.4% |
| Qwen3.8 Max | 88.7% | 90.8% |