Early Experiments in Accelerating Scientific Research with GPT-5

OpenAI published a paper of early case studies, in collaboration with universities and national laboratories, showing how GPT-5 can accelerate scientific research. The paper covers how GPT-5 performs literature reviews, accelerates complex calculations, and generates new proofs for unsolved problems across fields such as mathematics, physics, biology, and computer science.

Key research cases include:

• Biology: Analyzed unpublished charts to identify complex mechanisms of human immune cells in just minutes and suggested experimental directions.
• Mathematics: Provided new ideas for a decades-old problem posed by Paul Erdős, contributing to the completion of a proof.
• Algorithms and Optimization: Discovered error cases in robotics decision-making methods and improved upon existing results in optimization theory.

OpenAI for Science combines frontier models with specialized scientific tools such as simulation engines, protein databases, and computer algebra systems, enabling researchers to validate hypotheses faster.

This process is built on human-AI team collaboration. Scientists take the lead in setting the direction of research and verifying results, while GPT-5 maximizes the breadth and speed of research by connecting vast knowledge and suggesting mechanisms.