Math Reasoning Corpus · AI-MO
olympiads
Olympiad-level mathematical competition problems and solutions structured for mathematical reasoning models.
What olympiads is
AI-MO/olympiads is a curated reference dataset of mathematics competition problems and solutions drawn from national and international contests, including the International Mathematical Olympiad. Developed by AI-MO, the corpus is organised by competition to provide structured data for training and evaluating automated mathematical reasoning systems.
The dataset prioritises official solutions to ensure data quality. Entries are categorised by competition year, mathematical domain, question style, and difficulty, linking problems with step-by-step solutions.
How it was made
The dataset is compiled by gathering competition documents as raw PDF files and converting them into Markdown via Mathpix using the Numina data pipeline. Custom segmentation scripts then divide the converted text into structured JSON Lines files.
Each JSON entry includes the problem statement and its solution in LaTeX or Markdown format. Optional metadata fields capture the problem year, exam name, difficulty score, question style (proof, multiple-choice question, or math-word-problem), final answer value, and mathematical domain (such as Algebra, Geometry, Number Theory, Combinatorics, Calculus, Inequalities, or Logic and Puzzles). Diagrams referenced by placeholders in the text are stored in an accompanying Parquet file.
Key facts
- Publisher
- AI-MO
- Modality
- Document, text, and images
- Data formats
- PDF, Markdown (.md), JSON Lines (.jsonl), and Parquet (.parquet)
- Mathematical domains
- Algebra, Geometry, Number Theory, Combinatorics, Calculus, Inequalities, Logic and Puzzles, Other
- Question styles
- MCQ, proof, math-word-problem
- Creation date
- 21 November 2024
What you can use it for
- Mathematical reasoning training
- Supplying paired competition problems and complete solutions in LaTeX or Markdown to train models on advanced multi-step deduction.
- Model benchmarking across problem formats
- Evaluating the performance of automated reasoning systems on distinct task formats, including proofs, multiple-choice questions, and numerical-response word problems.
- Domain-specific performance analysis
- Measuring model accuracy across specialised subfields of mathematics, such as combinatorics, number theory, or geometry.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.Strengths and limits
Good at
- Prioritises official solutions to ensure mathematical accuracy and high presentation standards.
- Accommodates distinct problem types, including formal proofs, word problems with direct answers, and multiple-choice questions.
- Preserves visual contest information by associating in-text image references with an image Parquet file.
Watch out for
- Metadata fields such as year, difficulty rating, source, and exam name are optional and are not guaranteed to be present for every record.
- Problems containing diagrams require cross-referencing an external Parquet archive to inspect visual elements.
Questions
Can I look inside olympiads without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is olympiads under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
What formats are the problem statements and solutions provided in?
Problem statements and their solutions are formatted as LaTeX or Markdown strings inside segmented JSON Lines files.
Which mathematical subjects are represented in the dataset?
The dataset categorises items into Algebra, Geometry, Number Theory, Combinatorics, Calculus, Inequalities, Logic and Puzzles, and Other.
How are problem illustrations and figures handled?
Placeholders embedded in the problem text correspond directly to graphics stored in an accompanying images.parquet file.
What types of questions are contained in the corpus?
Entries are divided into multiple-choice questions (MCQ), formal proofs, and word problems that yield a specific final answer.
Written from olympiads's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.