OpenConjecture Dataset for Evaluating LLM Mathematical Reasoning
October 30, 2026
The OpenConjecture dataset provides a collection of randomly sampled mathematical conjectures from arXiv papers published since February 2026. Researchers used GPT models to test these conjectures via a reproducible protocol without problem-specific guidance to analyze solution characteristics and model capabilities.
HOW THIS AFFECTS YOU
●
researcherYou can use this dataset to evaluate how frontier models handle novel mathematical problems outside of curated benchmarks.