benchgap
Calibration

MMMU-Pro → MLVU (M-Avg)

MLVU (M-Avg) is estimated from MMMU-Pro with a linear curve fitted on 10 models measured on both: y = 0.4510·x + 0.5075, R² = 0.85, cross-validated error 2.2 pp. It is used for 22 estimates.

Estimated modelMMMU-ProMLVU (M-Avg)Source
GPT-6.1 Terra85.5%89.3%estimated ± 2.2 pp, medium confidence
GPT-5.6 Sol83.0%88.2%estimated ± 2.2 pp, medium confidence
GPT-5.581.2%87.4%estimated ± 2.2 pp, high confidence
GPT-5.6 Terra80.7%87.1%estimated ± 2.2 pp, high confidence
Hy478.6%86.2%estimated ± 2.2 pp, high confidence
GPT-5.6 Luna78.4%86.1%estimated ± 2.2 pp, high confidence
Grok 4.378.1%86.0%estimated ± 2.2 pp, high confidence
MiniMax M378.1%86.0%estimated ± 2.2 pp, high confidence
Pareto 26.978.0%85.9%estimated ± 2.2 pp, high confidence
Claude Opus 4.677.3%85.6%estimated ± 2.2 pp, high confidence
Gemma 4 31B76.9%85.4%estimated ± 2.2 pp, high confidence
GPT-5.4 mini76.6%85.3%estimated ± 2.2 pp, high confidence
Step 5 Preview76.0%85.0%estimated ± 2.2 pp, high confidence
Hy4 flash75.8%84.9%estimated ± 2.2 pp, high confidence
Gemma 4 26B A4B73.8%84.0%estimated ± 2.2 pp, high confidence
Interfaze Beta71.1%82.8%estimated ± 2.2 pp, high confidence
GPT-5.4 nano66.1%80.6%estimated ± 2.2 pp, high confidence
Gemma 4 E4B52.6%74.5%estimated ± 2.2 pp, high confidence
Qwen2.5-VL-32B49.5%73.1%estimated ± 2.2 pp, high confidence
Mistral Small 3.149.3%73.0%estimated ± 2.2 pp, high confidence
Gemma 4 E2B44.2%70.7%estimated ± 2.2 pp, medium confidence
LFM2.5-VL-3B30.5%64.5%estimated ± 2.2 pp, medium confidence

Fitted on

ModelMMMU-ProMLVU (M-Avg)
Claude Opus 4.570.6%81.7%
Gemini 3 Pro81.0%83.0%
GPT-5.279.5%85.6%
Kimi K2.578.5%85.0%
LFM3-VL-2B45.7%71.3%
Qwen3.5 397B79.0%86.7%
Qwen3.6-27B75.8%86.6%
Qwen3.6-35B-A3B75.3%86.2%
Qwen3.7 Plus79.0%87.4%
Qwen3.8 Max82.3%90.8%