benchgap
Calibration

ARC-AGI-2 → Vals MMLU-Pro

Vals MMLU-Pro is estimated from ARC-AGI-2 with a linear curve fitted on 24 models measured on both: y = 0.1142·x + 0.8059, R² = 0.69, cross-validated error 2.0 pp. It is used for 2 estimates.

Estimated modelARC-AGI-2Vals MMLU-ProSource
dots3-note Preview81.4%89.9%estimated ± 2.0 pp, high confidence
Gemini 3 Pro Deep Think45.1%85.7%estimated ± 2.0 pp, high confidence