{"paper":{"title":"IMProofBench: Benchmarking AI on Research-Level Mathematical Proof Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Adam Kurpisz, Aitor Iribar Lopez, Aleksandar Mijatovi\\'c, Alessio Bottini, Aluna Rizzoli, Ana Cannas da Silva, Ana-Maria Castravet, Baran Hashemi, Benjamin Doerr, Charles Vial, Chiara Meroni, Claudio Fontanari, Daniel Holmes, Daniel Platt, Daria Sakhanda, David Holmes, David Martins, David Mu\\~noz-Lahoz, Diaaeldin Taha, Dylan Possama\\\"i, Filippo Gaia, Gabriel Ribeiro, Georg Oberdieck, Gergely B\\'erczi, Henk van der Pol, Henry Liu, Ignacio Barros, Ingmar Metzler, Jasper Dekoninck, Jeremy Feusi, Jim Bryan, Jo\\~ao Camarneiro, Johannes Lengler, Johannes Schmitt, Josef Teichmann, Marc Roth, Martijn Kool, Martina J{\\o}rgensen, Martin M\\\"oller, Michel van Garrel, Niklas Canova, Patrick Schnider, Peter Gr\\\"unwald, Pieter Belmans, Pim Spelier, Raphael Appenzeller, Ra\\'ul S\\'anchez Gal\\'an, Richard P. Thomas, Robert Nowak, Ronald van Luijk, Samuel Mu\\~noz-Ech\\'aniz, Sergej Monavari, Stefan Kuhlmann, Steven Kelk, Tim Gehrunger, Timo de Wolff, Victor Jaeck, Yannik Schuler, Yuuji Tanaka, Zheming Sun","submitted_at":"2025-09-30T10:50:37Z","abstract_excerpt":"As the mathematical capabilities of large language models (LLMs) improve, it becomes increasingly important to evaluate their performance on research-level tasks at the frontier of mathematical knowledge. However, existing benchmarks are limited, as they focus solely on final-answer questions or high-school competition problems. To address this gap, we introduce IMProofBench, a private benchmark consisting of 77 peer-reviewed problems developed by expert mathematicians. Each problem requires a detailed proof and is paired with subproblems that have final answers, supporting both an evaluation "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.26076","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2509.26076/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}