Skip to content

Dataset · math-ai

AIME 2025 math benchmark

math-ai/aime25

Text-based competition problems from the 2025 American Invitational Mathematics Examination to evaluate model mathematical reasoning.

What's inside

A small text dataset with a volume of less than a thousand records (30 exam problems), containing tasks from the 2025 American Invitational Mathematics Examination (AIME). The dataset was published by the Math-AI team for standardized model testing. File formats and the presence of step-by-step solutions are not specified in the provided card fragment.

What you can do with it

  • Evaluating the ability of reasoning LLMs to solve complex mathematics
  • Comparing the quality of Chain-of-Thought generation before and after RLHF
  • Testing models for data leakage prevention when evaluating on fresh tasks
Size
up to 1K rows
Downloads this month
70.9K
Likes
49

How to load it

from datasets import load_dataset

ds = load_dataset("math-ai/aime25")