benchgap
Calibration

OfficeQA Pro → AA-MMMU-Pro

AA-MMMU-Pro is estimated from OfficeQA Pro with a Hill curve fitted on 8 models measured on both: y = 0.7677 + (1.2000 − 0.7677)·x^6.00 / (0.87218^6.00 + x^6.00), R² = 0.60, cross-validated error 2.9 pp. It is used for 5 estimates.

Estimated modelOfficeQA ProAA-MMMU-ProSource
Claude Fable 557.9%80.2%estimated ± 2.9 pp, high confidence
Claude Opus 4.866.2%83.7%estimated ± 2.9 pp, high confidence
Claude Sonnet 5.565.6%83.4%estimated ± 2.9 pp, high confidence
GLM-5.3-Flash62.4%81.9%estimated ± 2.9 pp, high confidence
Hy4 preview66.2%83.7%estimated ± 2.9 pp, high confidence