benchgap
Calibration

AIME26 → MATH-500

MATH-500 is estimated from AIME26 with a linear curve fitted on 5 models measured on both: y = 0.1364·x + 0.8468, R² = 0.69, cross-validated error 3.1 pp. It is used for 24 estimates.

Estimated modelAIME26MATH-500Source
A.X K297.1%97.9%estimated ± 3.1 pp, low confidence
Claude Opus 4.595.1%97.7%estimated ± 3.1 pp, medium confidence
Gemma 4 12B77.5%95.2%estimated ± 3.1 pp, medium confidence
GLM-595.8%97.7%estimated ± 3.1 pp, medium confidence
GLM-5.195.3%97.7%estimated ± 3.1 pp, medium confidence
GLM-5.299.2%98.2%estimated ± 3.1 pp, low confidence
Inkling97.1%97.9%estimated ± 3.1 pp, low confidence
Inkling-Small95.5%97.7%estimated ± 3.1 pp, medium confidence
K-EXAONE 2.092.3%97.3%estimated ± 3.1 pp, medium confidence
Kimi K2.696.4%97.8%estimated ± 3.1 pp, low confidence
Kimi K2.595.8%97.7%estimated ± 3.1 pp, medium confidence
Ling 3.0 Flash93.2%97.4%estimated ± 3.1 pp, medium confidence
LLaDA2.2-mini35.1%89.5%estimated ± 3.1 pp, low confidence
MAI-Thinking-194.5%97.6%estimated ± 3.1 pp, medium confidence
Muse Glimmer 30B94.7%97.6%estimated ± 3.1 pp, medium confidence
Qwen3.5 397B93.3%97.4%estimated ± 3.1 pp, medium confidence
Qwen3.6-27B94.1%97.5%estimated ± 3.1 pp, medium confidence
Qwen3.6-35B-A3B92.7%97.3%estimated ± 3.1 pp, medium confidence
Qwen3.6 Plus95.3%97.7%estimated ± 3.1 pp, medium confidence
Beam97.8%98.0%estimated ± 3.1 pp, low confidence
Solar Open 295.7%97.7%estimated ± 3.1 pp, medium confidence
Solar Pro 495.3%97.7%estimated ± 3.1 pp, medium confidence
ZAYA1-74B-Preview76.4%95.1%estimated ± 3.1 pp, medium confidence
ZAYA1-8B89.1%96.8%estimated ± 3.1 pp, medium confidence