benchgap
Calibration

HMMT Nov 2025 → AIME26

AIME26 is estimated from HMMT Nov 2025 with a Michaelis–Menten curve fitted on 10 models measured on both: y = 1.6199·x / (0.65234 + x), R² = 0.65, cross-validated error 1.6 pp. It is used for 1 estimate.

Estimated modelHMMT Nov 2025AIME26Source
Sarvam 30B74.2%86.2%estimated ± 1.6 pp, medium confidence

Fitted on

ModelHMMT Nov 2025AIME26
Claude Opus 4.593.3%95.1%
GLM-596.9%95.8%
GLM-5.194.0%95.3%
GLM-5.294.4%99.2%
Kimi K2.591.1%95.8%
MiniMax M2.781.0%89.8%
Qwen3.5 397B92.7%93.3%
Qwen3.6-27B90.7%94.1%
Qwen3.6-35B-A3B89.1%92.7%
Qwen3.6 Plus94.6%95.3%