FrontierScience is a new benchmark for evaluating AI’s expert-level scientific reasoning across physics, chemistry, and biology. It measures both Olympiad-style problem solving and real research tasks, helping track how well advanced models can support and accelerate scientific work.