Large language models (LLMs) have shown growing promise in solving research‑level problems, yet a systematic collection of such problems is still lacking and evaluating them on scientific challenges remains difficult. To address this gap we introduce OpenFrontiers, a cross‑disciplinary resource that combines a structured corpus of open research problems with a retrospective benchmark of recently solved mathematical problems. The corpus records 12,151 problems, including statements, context, progress and related metadata. The benchmark comprises 200 problems—110 resolved in 2026 and 90 still open—each equipped with a Lean 4 formal statement, validated by automated probes and reviewed by two independent assessors. A high‑quality subset (HQ‑50) is further filtered based on quality scores and other criteria. Our evaluation protocol grades formal proofs using two independent checkers inside an isolated sandbox under three conditions: closed‑book, frozen‑retrieval, and live‑web. Evaluating seven LLM agents on the HQ‑50 subset under a standardized budget reveals that closed‑book rediscovery is rare, at most 7 out of 40 resolved problems, while live‑web success largely reflects retrieval of published solutions. By linking a large research‑problem corpus to a formally graded evaluation subset with documented solutions, OpenFrontiers provides a foundation for measuring research‑level rediscovery and studying progress toward solving currently open scientific problems. The dataset and benchmark are available at https://huggingface.co/collections/FlippyDora/openfrontiers.
Review