Calibration

RealWorldQA → AA-MMMU-Pro

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

AA-MMMU-Pro is estimated from RealWorldQA with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (1.0606 − 0.0000)·x^6.00 / (0.73049^6.00 + x^6.00), R² = 0.78, cross-validated error 1.9 pp. It is used for 4 estimates.

Estimated model	RealWorldQA	AA-MMMU-Pro	Source
North Micro Vision Instruct	62.2%	29.3%	estimated ± 1.9 pp, low confidence
Qwen3.8 Max	88.0%	79.9%	estimated ± 1.9 pp, medium confidence
Qwen3.8-Omni-Flash	87.7%	79.5%	estimated ± 1.9 pp, medium confidence
Ternary Bonsai 2 27B	80.1%	67.4%	estimated ± 1.9 pp, low confidence