GPT-5 Pro Achieves 50% Improvement in Mathematical Bound, Spearheading OpenAI for Science's Inaugural Paper

San Francisco, CA – OpenAI for Science today announced the release of its inaugural paper, a comprehensive survey detailing the expanding applications of artificial intelligence across various scientific disciplines. The publication notably features novel mathematical results, including a significant 50% improvement in a convex optimization bound, derived by mathematician colleagues using the advanced AI model, GPT-5 Pro. This milestone underscores the initiative's commitment to accelerating scientific discovery through cutting-edge AI.

The OpenAI for Science initiative, led by Kevin Weil, Vice President of AI for Science, aims to establish AI as a foundational instrument for scientific research. Weil, formerly OpenAI's Chief Product Officer, has been instrumental in shaping this endeavor, which seeks to integrate frontier AI models with research tools to extend human curiosity and drive breakthrough discoveries. The initiative is actively recruiting world-class academics to collaborate with OpenAI researchers.

Alex Lupsasca, an Assistant Professor of Physics and Mathematics at Vanderbilt University, has been identified as the first academic to join this pioneering initiative. Lupsasca, a theoretical physicist, has previously attested to GPT-5 Pro's capacity to solve complex physics problems in minutes, tasks that would typically require days for human researchers. His involvement highlights the practical application of AI in high-level scientific inquiry.

GPT-5 Pro is lauded for its deep reasoning, conceptual search, and formal computational capabilities, enabling it to act as a knowledgeable research partner. Beyond the mathematical bound improvement, the model has been instrumental in generating correct proofs and assisting physicists in rediscovering hidden symmetry structures. These early contributions signify a meaningful shift in how AI can directly contribute to fundamental science under expert oversight.

The newly released survey outlines a broad spectrum of AI uses in science, positioning GPT-5 Pro as a "new threshold" in AI-assisted research. OpenAI for Science emphasizes that such AI systems are designed to empower researchers, allowing them to explore more ideas and test hypotheses faster, ultimately compressing decades of discovery into years. The initiative is expected to continue publishing findings and expanding its collaborative efforts with the scientific community.