# hendrycks/math

The MATH Dataset (NeurIPS 2021)

Repository: https://github.com/hendrycks/math
Canonical: https://ross.abutalabs.com/products/hendrycks-math
Language: Python
License: MIT
License Family: permissive
Topics: math, reasoning, logic
Last push: 2025-09-06T18:25:04+00:00

## Health v2 (maintenance only)
Score: 50/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 40, release rhythm 35, longevity 100
- inputs: {"age_days": 2030, "days_push": 361, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1386, forks 116 (observed 2026-08-28T04:04:35.043592+00:00)

## What it is
The MATH Dataset is a benchmark of 12,500 competition mathematics problems with full step-by-step solutions, used to measure mathematical problem solving in machine learning models. The repository provides dataset loaders and evaluation code accompanying the NeurIPS 2021 paper.

## Use cases
- evaluate llm math reasoning ability
- benchmark models on competition math problems
- get a dataset of step-by-step math solutions
- test mathematical problem solving of language models
- load the MATH dataset for research
- compare model performance on hard math word problems

## When to choose
- you need a rigorous benchmark for mathematical reasoning in LLMs
- you want competition-level math problems with worked solutions
- you are reproducing or comparing against published MATH benchmark results

## When to avoid
- you need elementary arithmetic drills or simple word problems
- you want a training corpus rather than an evaluation benchmark (though AMPS exists for pretraining)
- you need automated grading of arbitrary student answers

## Facets
- artifact type: dataset
- maturity: stable
- function: machine-learning, math, benchmarking, data-science
- domain: mathematics, machine-learning, education
- platform: python
- tags: math-dataset, benchmark, reasoning, evaluation, competition-math, neurips-2021, natural-language-processing

## Member repositories
- hendrycks/math (main) score 50

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:35.043592+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:39:54.355817+00:00, confidence not recorded.
  - readme: https://github.com/hendrycks/math (fetched 2026-08-28T04:04:35.043592+00:00, sha ea2a34912f19)
- Data as of 2026-08-30T08:39:29.467469+00:00.