AMC12: American Mathematics Competition Problems from 2000 to 2025

A structured dataset derived from the AMC 12 (American Mathematics Competitions), designed for LLM training, evaluation, and reinforcement learning on mathematical reasoning tasks. It contains all AMC 12 problems from 2000 to 2025, making it one of the most complete AMC12 datasets available for research. The dataset was created by edev2000 and last updated on March 17, III.

Use Cases

• Training language models on mathematical problem-solving based on the structured competition problems.
• Evaluating model performance on high school-level mathematics based on the AMC 12 exam format.
• Reinforcement learning for step-by-step mathematical reasoning based on the multiple-choice question structure.
• Benchmarking AI systems on competition-level mathematics based on the 25-question exam format.

Strengths

• Contains all AMC 12 problems from a 25-year time range (2000 to 2025).
• Designed specifically for LLM training, evaluation, and reinforcement learning research.
• Based on a standardized, high school-level mathematics competition with a defined 25-question, 75-minute format.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal and demographic bias inherent to the source competition.

Provenance

Source

American Mathematics Competitions (AMC 12).

Collection Method

Derived from the competition problems.

Time Range

2000 to 2025.

Freshness

Last updated 2026-03-17 19:06:07; freshness should be verified.

Geography

United States (based on competition name and region tag).