OpenAI has launched FrontierScience, a new benchmark to assess expert-level AI scientific reasoning across physics, chemistry and biology, as models like GPT-5 increasingly support real research.