FrontierScience is a dataset for evaluating inference and scientific research tasks, released by OpenAI in 2025. Related research papers are published in FrontierScience.evaluating AI's ability to perform expert-level scientific tasksThe aim is to systematically evaluate the capabilities of large models in expert-level scientific reasoning and research sub-tasks.

This dataset employs a design mechanism of "expert creation + two-layer task structure + automatic scoring mechanism," and is divided into two subsets, corresponding to two types of abilities: closed-ended precise reasoning and open-ended scientific research reasoning.

• The Olympiad dataset was originally designed by medal winners and national team coaches from the International Physics, Chemistry and Biology Olympiads. The difficulty of the questions is comparable to top international competitions such as IPhO, IChO and IBO. It focuses on short-answer reasoning tasks and requires the model to output a single numerical value, an algebraic expression or a biological term that can be fuzzily matched, in order to ensure the verifiability of the results and the stability of the automatic evaluation.

• The Research Dataset is written by PhD students, postdoctoral fellows, professors, and other active researchers. The questions simulate sub-problems that may be encountered in real scientific research, covering the three major fields of physics, chemistry, and biology. Each question is accompanied by a fine-grained score of 10 points to evaluate the model's performance in several key aspects, including modeling assumptions, reasoning paths, and intermediate conclusions, in addition to the correctness of the answer.