Title: Reasoning4Sciences: Connecting Reasoning Language Models to Every Scientific Domain

Abstract:

Although Reasoning Language Models (RLMs) are swiftly becoming indispensable assets for scientific inquiry, their influence remains heavily skewed toward "hard science" disciplines. This uneven uptake is fostering a growing disparity in research output, as other scientific branches lag behind in adopting these technologies. In this survey, we present the inaugural holistic examination of RLM integration across 28 distinct scientific fields, utilizing the classification system established by the European Research Council (ERC). Our scope encompasses the Social Sciences and Humanities, Physical Sciences and Engineering, and Life Sciences.

We analyze the trajectories of RLM development, evaluation, and practical application within these varied sectors. Additionally, we propose a maturity-focused assessment framework derived from the existing landscape of domain-specific development and evaluation tools. This framework exposes significant imbalances in RLM maturity levels, a trend that intensifies when the analysis is restricted to resources that are publicly accessible. Lastly, we identify emerging implementation paradigms that are gaining traction across disciplines, while also outlining present obstacles and prospective pathways to facilitate broader RLM adoption throughout the scientific community.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC