Calibration

MMLU-Pro → SuperGPQA

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

SuperGPQA is estimated from MMLU-Pro with a inverse Michaelis–Menten curve fitted on 17 models measured on both: y = 1.05013·(x − 0.0582) / (0.0582 + 2.0000 − x), R² = 0.44, cross-validated error 12.7 pp. It is used for 1 estimate.

Estimated model	MMLU-Pro	SuperGPQA	Source
Solar Open 2	86.2%	70.6%	estimated ± 12.7 pp, low confidence