Math Reasoning research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

MATH-500 (held-out)

alpha

4.79Accuracy Delta (pp)

May 15, 2026

HMMT25 Nov.

MOPD

15.83Mean@8

May 14, 2026

HMMT Feb 25

MOPD

18.5Mean @8

May 14, 2026

AIME 2024

MOPD

28.54Mean@8 Score

May 14, 2026

AIME 2022–2024

GRPO + WeMask (TF)

9.27Accuracy

May 12, 2026

VoxEval Average

Qwen2.5-7B-Instruct

73.1Accuracy

May 8, 2026

VoxEval College

Qwen2.5-7B-Instruct

63Accuracy

May 8, 2026

VoxEval High School

Qwen2.5-7B-Instruct

73.1Accuracy

May 8, 2026

VoxEval Elementary

Qwen2.5-7B-Instruct

83.2Accuracy

May 8, 2026

Math Reasoning Multimodal Benchmarks

Teacher

47.2Math Vision Accuracy

May 6, 2026

AIME25

ESamp

31.7Pass@16

Apr 29, 2026

AIME '24

IMH

9.5Pass@1

Apr 2, 2026

AIME 2025 (test)

GEPA

49.06Mean Accuracy

Feb 26, 2026

DROP numerical problems (val)

Least-to-Most

82.45Non-football Accuracy

Feb 27, 2026

AQuA-RAT

59.06Accuracy

May 28, 2026

MATH-500

WINO+

34.4Accuracy

May 19, 2026

GSM8K

WINO+

76.58Accuracy

May 19, 2026

AIME 2025 (test)

Merged (α=0.2)

50Pass@1 Rate

Apr 14, 2026

AMC (out-of-distribution)

CalibRL

32.08Accuracy

Feb 26, 2026

Math Reasoning 1.5B model (val)

Execution-Guided Search

69.4Validation Accuracy

Feb 26, 2026

Large-scale model pool Math Reasoning 15 LLMs

RouteMoA

73.3Accuracy

Feb 26, 2026

HMMT November 2025 edition (test)

ETGPO

4,025Optimization Token Usage

Feb 26, 2026

AIME 2025 edition (test)

ETGPO

4,377Optimization Tokens

Feb 26, 2026

HMMT (test)

ETGPO

3,295Optimization-phase Token Usage (thousands)

Feb 26, 2026

AIME (test)

ETGPO

3,098Token Usage (Optimization Phase, Thousands)

Feb 26, 2026