PolyMath is a curated dataset of 11,090 high-difficulty mathematical problems designed for training reasoning models. It was created by AIMO-Corpus for the AIMO Math Corpus Prize and was last updated on February 9, 2026. The dataset addresses noise and usability issues found in other math datasets.
Use Cases
- Training mathematical reasoning models based on high-difficulty problems.
- Benchmarking AI performance on competition-level math tasks.
- Improving proof generation models based on curated, high-quality problems.
- Studying the structure of complex mathematical reasoning tasks.
Strengths
- Contains 11,090 specifically curated problems.
- Focuses on high-difficulty problems sourced from official competition PDFs.
- Addresses noise and usability issues identified in other datasets.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- AIMO-Corpus
- Collection Method
- Data scraping from official competition PDFs.
- Freshness
- Last updated 2026-02-09 22:36:35; freshness should be verified.