SorryDB: Can AI Provers Complete Real-World Lean Theorems?

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Letson, Austin, Sarra, Leopoldo, Poiroux, Auguste, Dressler, Oliver, Lezeau, Paul, Aranha, Dhyan, Pu, Frederick, Hill, Aaron, Hidalgo, Miguel Corredera, Berman, Julian, Tsoukalas, George, Taelman, Lenny
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866912939264442368
author Letson, Austin
Sarra, Leopoldo
Poiroux, Auguste
Dressler, Oliver
Lezeau, Paul
Aranha, Dhyan
Pu, Frederick
Hill, Aaron
Hidalgo, Miguel Corredera
Berman, Julian
Tsoukalas, George
Taelman, Lenny
author_facet Letson, Austin
Sarra, Leopoldo
Poiroux, Auguste
Dressler, Oliver
Lezeau, Paul
Aranha, Dhyan
Pu, Frederick
Hill, Aaron
Hidalgo, Miguel Corredera
Berman, Julian
Tsoukalas, George
Taelman, Lenny
contents We present SorryDB, a dynamically-updating benchmark of open Lean tasks drawn from 78 real world formalization projects on GitHub. Unlike existing static benchmarks, often composed of competition problems, hillclimbing the SorryDB benchmark will yield tools that are aligned to the community needs, more usable by mathematicians, and more capable of understanding complex dependencies. Moreover, by providing a continuously updated stream of tasks, SorryDB mitigates test-set contamination and offers a robust metric for an agent's ability to contribute to novel formal mathematics projects. We evaluate a collection of approaches, including generalist large language models, agentic approaches, and specialized symbolic provers, over a selected snapshot of 1000 tasks from SorryDB. We show that current approaches are complementary: even though an agentic approach based on Gemini Flash is the most performant, it is not strictly better than other off-the-shelf large-language models, specialized provers, or even a curated list of Lean tactics.
format Preprint
id arxiv_https___arxiv_org_abs_2603_02668
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SorryDB: Can AI Provers Complete Real-World Lean Theorems?
Letson, Austin
Sarra, Leopoldo
Poiroux, Auguste
Dressler, Oliver
Lezeau, Paul
Aranha, Dhyan
Pu, Frederick
Hill, Aaron
Hidalgo, Miguel Corredera
Berman, Julian
Tsoukalas, George
Taelman, Lenny
Artificial Intelligence
Machine Learning
We present SorryDB, a dynamically-updating benchmark of open Lean tasks drawn from 78 real world formalization projects on GitHub. Unlike existing static benchmarks, often composed of competition problems, hillclimbing the SorryDB benchmark will yield tools that are aligned to the community needs, more usable by mathematicians, and more capable of understanding complex dependencies. Moreover, by providing a continuously updated stream of tasks, SorryDB mitigates test-set contamination and offers a robust metric for an agent's ability to contribute to novel formal mathematics projects. We evaluate a collection of approaches, including generalist large language models, agentic approaches, and specialized symbolic provers, over a selected snapshot of 1000 tasks from SorryDB. We show that current approaches are complementary: even though an agentic approach based on Gemini Flash is the most performant, it is not strictly better than other off-the-shelf large-language models, specialized provers, or even a curated list of Lean tactics.
title SorryDB: Can AI Provers Complete Real-World Lean Theorems?
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2603.02668