NuminaMath 1.5: 900,000 Competition-Level Math Problems with CoT Solutions

AI-MO released this collection of approximately 900,000 competition-level math problems in early 2026 for model post-training. It features Chain of Thought (CoT) solutions derived from Chinese high school exercises and international mathematics olympiads.

Use Cases

• Fine-tuning language models for mathematical reasoning using the Chain of Thought solution format
• Benchmarking model performance on international math olympiad problems
• Training text generation models to solve Chinese high school math exercises

Strengths

• 900,000 competition-level records
• Chain of Thought (CoT) formatted solutions
• Sourced from diverse international and Chinese olympiad exams

Limitations

• Potential for noise or formatting errors due to PDF-to-text extraction
• Geographic bias towards US and Chinese educational systems

Provenance

Source

AI-MO

Collection Method

Scraped and processed from online exam paper PDFs

Freshness

Last updated January 2026

Geography

International (US and China)