{
  "id": "math-benchmark",
  "code": "PTL-0198",
  "term": "MATH",
  "aliases": [
    "MATH dataset",
    "MATH-500"
  ],
  "category": "benchmarks",
  "definition": "MATH is a benchmark of 12,500 competition mathematics problems, spanning seven subjects and five difficulty levels, each with a full step-by-step solution.",
  "description": "It is much harder than grade-school arithmetic and was long a key measure of mathematical reasoning. A 500-problem subset, MATH-500, created for research on process supervision, is the version most often reported today.",
  "example": null,
  "broader": [
    "benchmark"
  ],
  "narrower": [],
  "related": [
    "gsm8k",
    "aime",
    "process-reward-model"
  ],
  "introduced": 2021,
  "added": "2026-10-10",
  "sources": [
    {
      "title": "Measuring Mathematical Problem Solving With the MATH Dataset",
      "authors": "Hendrycks et al.",
      "year": 2021,
      "url": "https://arxiv.org/abs/2103.03874"
    },
    {
      "title": "Let's Verify Step by Step",
      "authors": "Lightman et al.",
      "year": 2023,
      "url": "https://arxiv.org/abs/2305.20050"
    }
  ],
  "url": "https://protologue.com/t/math-benchmark/",
  "citation": "Protologue. (2026). MATH. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0198). https://protologue.com/t/math-benchmark/"
}