Competition Math · AI-MO
aimo-validation-aime
A validation benchmark of 90 recent AIME competition mathematics problems with integer solutions.
What aimo-validation-aime is
aimo-validation-aime is an evaluation dataset created by AI-MO to benchmark mathematical reasoning systems. It was compiled as an internal validation suite for the Artificial Intelligence Mathematical Olympiad (AIMO) progress prize competition.
Each problem follows a 15-question structure requiring an integer answer, spanning difficulty levels from 5 to 9.
To help researchers reliably evaluate mathematical models without data contamination, the collection is limited to competitions held after 2021, minimising potential overlap with widely used training corpora such as the MATH dataset.
How it was made
The records include both Part I and Part II examinations across those three years, yielding 90 questions in total.
Each entry contains three fields: the original problem statement, a reference solution taken from the forum discussions containing the final answer formatted in a LaTeX box, and the source URL pointing to the relevant AoPS wiki entry.
Key facts
- Total problems
- 90
- Difficulty range
- Levels 5–9
- Data format
- Parquet
- Input modality
- Text
- Fields
- problem, solution, url
- Release date
- 9 July 2024
What you can use it for
- Mathematical reasoning evaluation
- Assessing the multi-step problem-solving capabilities of automated reasoning systems on advanced competition mathematics.
- Contamination-free validation
- Validating models trained on the MATH benchmark against post-2021 contest items to prevent memorisation and data leakage.
- AIMO challenge benchmarking
- Providing a reference validation set calibrated directly for the format and difficulty of the AIMO progress prize.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from AI-MO/aimo-validation-aime at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 90 rows of default/train
answer
string_text- min
- 2
- median
- 3
- mean
- 2.99
- max
- 3
id
int- min
- 0
- median
- 44.50
- mean
- 44.50
- max
- 89
problem
string_text- min
- 114
- median
- 289.50
- mean
- 341.62
- max
- 938
solution
string_text- min
- 392
- median
- 4,590
- mean
- 5333.51
- max
- 13,820
url
string_text- min
- 77
- median
- 78
- mean
- 77.90
- max
- 79
Strengths and limits
Good at
- It uses post-2021 problems specifically to avoid data contamination from the standard MATH training set.
- Every problem is paired with a forum-sourced explanation and a clearly identified integer answer in LaTeX boxed format.
- It covers high difficulty tiers (levels 5–9) across three consecutive years of AIME examinations.
Watch out for
- The corpus is limited in scale to 90 total problems.
- Reference solutions represent a single forum-suggested solution per problem rather than exhaustive alternative proofs.
- The scope is restricted exclusively to the 15-question integer-answer format of AIME competitions.
Questions
Can I look inside aimo-validation-aime without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is aimo-validation-aime under?
Its authors declare apache-2.0. Read the terms on the original page before reusing the data.
What data fields are present in the dataset?
The records contain three columns: the original problem text, a solution containing the boxed integer answer, and the URL to the AoPS wiki page.
What answer format do the problems require?
In accordance with standard AIME competition rules, each problem requires an integer answer.
Written from aimo-validation-aime'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.