MATH (Hendrycks)
12,500 competition mathematics problems across 7 subjects (algebra, calculus, number theory, etc.) with step-by-step solutions. The key benchmark for evaluating mathematical reasoning in LLMs — requires multi-step problem-solving and symbolic manipulation.
from datasets import load_dataset
ds = load_dataset("hendrycks/competition_math")Preview a sample row
{
"problem": "Find the integer $n$, $0 \\le n \\le 11$, such that $n \\equiv 10389 \\pmod{12}$.",
"solution": "Since $10389 = 865 \\cdot 12 + 9$, the remainder when 10389 is divided by 12 is $\\boxed{9}$.",
"level": "Level 1",
"type": "Number Theory"
}Frequently asked questions
Can I use MATH (Hendrycks) commercially?
Yes — MATH (Hendrycks) is released under MIT, a permissive license that allows commercial use, including training models you ship in a product. Check the dataset card for attribution requirements before release.
How much data does MATH (Hendrycks) contain, and do I need all of it?
MATH (Hendrycks) contains 12,500 Problems. It is an evaluation benchmark, so it is used in full to measure models — never mix it into training data, or your benchmark scores become meaningless.
What is MATH (Hendrycks) best used for?
Benchmarking competition math ability. It belongs to the Evaluation & Benchmarks section of our dataset hub, where you'll find alternatives and complementary sets.