Skip to content
← Models
Open weightsProject Numina/10 trustVerified

NuminaMath 72B

The open model that won the first AI Math Olympiad progress prize — $131,072 of proof.

Visit official page →

About

NuminaMath-72B is the model behind Team Numina's win of the first AI Mathematical Olympiad Progress Prize in July 2024 — a $10M XTX Markets competition testing whether AI can solve olympiad-level math with no internet, on open models only. Built by fine-tuning Qwen2-72B on NuminaMath-CoT, the largest public math reasoning dataset ever released (860,000 problem-solution pairs), it solved 29 of 50 competition problems using self-consistency tool-integrated reasoning.

Project Numina — backed by a €3M XTX Markets grant and supported by Hugging Face, Mistral AI, and Answer.ai — released the model, dataset, and training scripts openly, making the win a landmark for open science: the first AIMO prize went to a fully open-source solution, not a closed lab.

Honest caveats: it is a competition-math specialist tuned for the AIMO answer format (integers 0–99999), not a general assistant. The 72B size needs serious hardware. Check the Hugging Face model card for the exact license before commercial use. Later AIMO editions were won by larger efforts, but NuminaMath remains the proof that open models can win at elite math.

Specs

Context window

Pricing

Free quota

API access

Weights

Open weights

Released

July 22, 2024

Benchmarks

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

math-reasoningcompetition-matheducation

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 61; Trending score 76/100 from researcher assessment.

Sources