cff-version: 1.2.0
message: "If you use this dataset, please cite it using the metadata below."
title: "AIME++ Sample"
type: dataset
authors:
  - name: "Ulam AI"
version: 0.1.0
date-released: 2026-08-26
repository-code: "https://huggingface.co/datasets/ulamai/AIME-Plus-Plus"
url: "https://ulam.ai/"
license: MIT
abstract: >-
  A sample of 157 exact-answer mathematical reasoning problems spanning AIME,
  AIME Hard, AIME-Graduate, and AIME-Researcher difficulty tiers. Every answer
  is an integer from 0 through 999 and supports deterministic exact-match
  evaluation.
keywords:
  - mathematical reasoning
  - benchmark
  - exact-match evaluation
  - verifiable rewards