Competition Math · AI-MO
aimo-validation-amc
A validation benchmark of 83 reformatted AMC 12 competition problems with integer answers for mathematical reasoning.
What aimo-validation-amc is
aimo-validation-amc is a curated mathematical evaluation dataset developed by AI-MO. Created as an internal validation suite for the AI Mathematical Olympiad (AIMO) Progress Prize competition, it evaluates the mathematical reasoning capabilities of automated systems across topics such as algebra, geometry, and number theory spanning difficulty levels 1 to 5.
The dataset addresses the challenge of data contamination in machine learning benchmarks. By sourcing problems exclusively from the 2022 and 2023 American Mathematics Competitions (AMC 12), it avoids temporal overlap with older corpora, specifically the commonly used MATH training set which relies on pre-2021 material.
To match the target competition environment where models must produce exact numerical values rather than selecting from a list, the original multiple-choice questions have been systematically converted to require single integer outputs.
How it was made
The problems were collected from the Art of Problem Solving (AoPS) wiki pages covering the 2022 and 2023 AMC 12 competitions. The source contests originally presented problems in a multiple-choice format with five options.
During dataset creation, each question was adapted to require an integer answer, such as requesting the sum of the numerator and denominator of a simplified fraction. Questions whose formulations could not be modified to yield an integer answer were rejected. The final dataset contains 83 examples stored across three attributes: the modified problem prompt, the adapted integer solution, and the original AoPS source URL.
Key facts
- Publisher
- AI-MO
- Total instances
- 83
- Source competitions
- AMC 12 (2022 and 2023)
- Subject coverage
- Algebra, geometry, number theory
- Difficulty range
- Levels 1 to 5
- Data format
- Parquet, tabular text
- Fields
- problem, answer, url
- Target output type
- Integer
What you can use it for
- AIMO Progress Prize Validation
- Benchmarking competition mathematics models against recent high-school competition problems adapted to the AIMO integer output format.
- Contamination-Free Evaluation
- Assessing mathematical reasoning models trained on pre-2021 datasets without overlap from the standard MATH training corpus.
- Mathematical Reasoning Assessment
- Evaluating algorithmic performance in algebra, geometry, and number theory problems rated across difficulty tiers 1 to 5.
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-amc at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 83 rows of default/train
answer
float- min
- -4
- median
- 12
- mean
- 222.54
- max
- 8,178
id
int- min
- 0
- median
- 23
- mean
- 23.67
- max
- 49
problem
string_text- min
- 72
- median
- 264
- mean
- 291.33
- max
- 941
url
string_text- min
- 78
- median
- 79
- mean
- 78.61
- max
- 79
Strengths and limits
Good at
- Eliminates overlap with pre-2021 mathematical datasets such as the MATH training set by using only post-2021 AMC 12 competitions.
- Reformulates multiple-choice questions into integer-answer queries to properly evaluate generative problem-solving rather than option guessing.
- Includes AoPS source URLs for direct verification of the original questions and standard solutions.
Watch out for
- Contains a relatively small sample size of 83 problems.
- Excludes questions from the source competitions that could not be cleanly converted to yield integer answers.
- Restricted to AMC 12 questions from 2022 and 2023, omitting other competition tiers or subsequent years.
Questions
Can I look inside aimo-validation-amc 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-amc under?
Its authors declare apache-2.0. Read the terms on the original page before reusing the data.
What is the primary purpose of this dataset?
It serves as an evaluation set for assessing model performance on competition mathematics under conditions matching the AIMO Progress Prize.
Why were only problems from 2022 and 2023 selected?
Selecting problems after 2021 ensures that models trained on standard benchmarks, such as the MATH training set, have not previously encountered the test problems during training.
How do these questions differ from original AMC 12 questions?
Original AMC 12 problems offer five multiple-choice options, whereas this dataset modifies the questions to require an integer answer and omits problems that could not be adapted.
What attributes are present in each dataset row?
Each row contains three attributes: 'problem' for the modified problem text, 'answer' for the adapted integer answer, and 'url' linking to the corresponding Art of Problem Solving wiki page.
Written from aimo-validation-amc'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.