benchgap
Calibration

MathVision → MMVU

MMVU is estimated from MathVision with a Michaelis–Menten curve fitted on 5 models measured on both: y = 2.0000·x / (1.41534 + x), R² = 0.66, cross-validated error 2.9 pp. It is used for 14 estimates.

Estimated modelMathVisionMMVUSource
Claude Opus 4.574.3%68.8%estimated ± 2.9 pp, low confidence
Gemini 3 Pro86.6%75.9%estimated ± 2.9 pp, medium confidence
Gemma 4 12B79.7%72.1%estimated ± 2.9 pp, low confidence
GPT-5.283.0%73.9%estimated ± 2.9 pp, low confidence
Kimi K2.687.4%76.4%estimated ± 2.9 pp, medium confidence
Kimi K394.3%80.0%estimated ± 2.9 pp, medium confidence
Qwen3.5 397B88.6%77.0%estimated ± 2.9 pp, medium confidence
Qwen3.6 Plus88.0%76.7%estimated ± 2.9 pp, medium confidence
Qwen3.7 Plus90.3%77.9%estimated ± 2.9 pp, medium confidence
Qwen3.8-27B90.0%77.7%estimated ± 2.9 pp, medium confidence
Qwen3.8-Flash-Next90.6%78.1%estimated ± 2.9 pp, medium confidence
Qwen3.8-Omni-Flash91.8%78.7%estimated ± 2.9 pp, medium confidence
Seed 2.1 Pro92.6%79.1%estimated ± 2.9 pp, medium confidence
Seed 2.1 Turbo90.1%77.8%estimated ± 2.9 pp, medium confidence