II.
Benchmark overview
Reference · livebenchmark:math
MATH overview
The MATH benchmark (Hendrycks et al., 2021) is 12,500 competition mathematics problems drawn from AMC, AIME, and similar contests, with full step-by-step solutions.
Attributes
displayName
MATH
homepageUrl
kind
math
targetsKind
ModelVersion
description
The MATH benchmark (Hendrycks et al., 2021) is 12,500 competition
mathematics problems drawn from AMC, AIME, and similar contests,
with full step-by-step solutions.
Outgoing edges
None.
Incoming edges
belongs_to_benchmark1
- test-set:math-test·TestSetMATH test split
bounds_subject1
- scope-boundary:math.scope·ScopeBoundary
for_benchmark3
- eval-run:math.deepseek-r1.2025-01·EvalRun
- eval-run:math.o3.2025-04·EvalRun
- eval-run:math.gpt-5.2025-08·EvalRun
scored_against3
- eval-result:math.deepseek-r1.001·EvalResult
- eval-result:math.o3.001·EvalResult
- eval-result:math.gpt-5.001·EvalResult