Datasets:

AI-MO
/

aops

dataset_info:
  features:
    - name: problem
      dtype: string
    - name: solution
      dtype: string
    - name: candidates
      sequence: string
    - name: tags
      sequence: string
    - name: metadata
      struct:
        - name: answer_score
          dtype: int64
        - name: boxed
          dtype: bool
        - name: end_of_proof
          dtype: bool
        - name: n_reply
          dtype: int64
        - name: path
          dtype: string
  splits:
    - name: train
      num_bytes: 298608789
      num_examples: 80661
  download_size: 140630996
  dataset_size: 298608789
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

AoPS: Art of Problem Solving Competition Mathematics

Dataset Description

This dataset is a collection of 80,661 competition mathematics problems and solutions obtained from the Art of Problem Solving (AoPS) community wiki and forums. It covers a wide range of mathematical contests and olympiads, including problems from events such as AIME, BAMO, IMO, and various national and memorial competitions.

The dataset was curated by AI-MO (Project Numina), an initiative focused on building AI systems capable of mathematical reasoning at the olympiad level.

Dataset Structure

Fields

Column	Type	Description
`problem`	`string`	The mathematical problem statement, typically formatted in LaTeX.
`solution`	`string`	A solution or proof for the problem. May be empty for some entries.
`candidates`	`list[string]`	Alternative or candidate solutions contributed by the community.
`tags`	`list[string]`	Metadata tags indicating the origin, contest name, and year (e.g., `"origin:aops"`, `"2022 AIME Problems"`).
`metadata`	`dict`	Additional metadata about the problem (see below).

Metadata Fields

Field	Type	Description
`answer_score`	`int64`	Community score or rating of the answer.
`boxed`	`bool`	Whether the answer contains a boxed final result (e.g., `\boxed{42}`).
`end_of_proof`	`bool`	Whether the solution includes a complete proof ending.
`n_reply`	`int64`	Number of community replies or comments on the problem thread.
`path`	`string`	Source path in the AoPS collection (e.g., `Contest Collections/2022 Contests/...`).

Splits

Split	Examples
`train`	80,661

Example

{
    "problem": "Let $ABC$ be an acute triangle with altitude $AD$ ($D \\in BC$). The line through $C$ parallel to $AB$ meets the perpendicular bisector of $AD$ at $G$. Show that $AC = BC$ if and only if $\\angle AGC = 90°$.",
    "solution": "...",
    "candidates": ["..."],
    "tags": ["origin:aops", "2022 Contests", "2022 3rd Memorial \"Aleksandar Blazhevski-Cane\""],
    "metadata": {
        "answer_score": 130,
        "boxed": false,
        "end_of_proof": true,
        "n_reply": 3,
        "path": "Contest Collections/2022 Contests/2022 3rd Memorial .../2759376.json"
    }
}

Topic Coverage

Problems span a broad range of competition mathematics topics, including:

Geometry -- triangle properties, cyclic quadrilaterals, angle chasing
Number Theory -- divisibility, modular arithmetic, Diophantine equations
Algebra -- inequalities, polynomials, functional equations
Combinatorics -- counting, graph theory, board coloring problems

Usage

from datasets import load_dataset

dataset = load_dataset("AI-MO/aops")

# Access a problem
print(dataset["train"][0]["problem"])
print(dataset["train"][0]["solution"])

Intended Use

Training and evaluating mathematical reasoning models
Benchmarking LLMs on competition-level mathematics
Studying solution quality and problem difficulty distributions
Building retrieval-augmented generation (RAG) systems for math tutoring

Source

All problems and solutions originate from the Art of Problem Solving community.