A Primer in Post-Training Reasoning Data: What We Know About How It Works
2026-06-01 • Computation and Language
Computation and LanguageArtificial Intelligence
AI summaryⓘ
The authors looked at many studies about improving large AI models by training them after their initial creation, especially focusing on reasoning skills. They noticed that information used during this extra training, called reasoning data, is very important but scattered across many different papers. They organized the research into four main questions: what types of data are used, why the data helps, how it's made, and how it grows with model size. This helps future researchers understand and improve how reasoning data is used in training AI models.
post-trainingreasoning datalarge language modelsreinforcement learningreward modelsbenchmarksdataset constructionscaling lawsfine-tuning
Authors
Yaoming Li, Guangxiang Zhao, Qilong Shi, Lin Sun, Xiangzheng Zhang, Tong Yang
Abstract
Post-training has become a primary driver of recent progress in large reasoning models, and reasoning data are often the key variable determining whether this stage succeeds. Work on post-training reasoning data has grown rapidly, yet this literature remains scattered across dataset papers, reinforcement-learning recipes, reward-model studies, benchmarks, and frontier system reports. This paper is the first primer to synthesize over 150 key public studies and system reports on post-training reasoning data. We organize the field around four questions: what data objects exist, what makes them useful, how they are constructed, and how they scale. Together, this organization provides an attribution framework for future reasoning-data releases and post-training recipes.