The exponential pace of artificial intelligence development is poised to shift its axis of maximum impact, moving from the revolutionary gains seen in software engineering to a profound acceleration in scientific discovery, according to a leading figure at OpenAI. Kevin Weil, the company’s Vice President for Science, has articulated a bold vision, stating that “2026 will be for science what 2025 was for software engineering,” suggesting a coming year of step-change productivity that could reshape the entire scientific landscape. His projection is rooted in the proven capabilities of advanced models like GPT-5 and the rapidly escalating adoption of AI tools by research communities worldwide, pointing toward a transformative period where AI moves beyond an intelligent assistant to a true scientific collaborator.

The benchmark for this forthcoming scientific revolution is the dramatic productivity increase experienced by the software development sector in the prior period. The approximate timeframe of 2025 saw a fundamental shift as AI models progressed from sophisticated question-answer tools to true agents capable of executing complex, real-world tasks. OpenAI’s own GPT-5 model was heralded as an “incredible coding model,” delivering highly agentic capabilities that allowed developers to automate intricate workflows, follow complex instructions, and integrate disparate services without losing context.[1][2] This wave of tooling fundamentally amplified human capacity, with Weil noting the potential for AI coding models to expand the global population that "effectively know how to code" from a mere 30 million to as many as three billion people, thereby democratizing software creation and leading to an explosion of new personalized and specialized applications.[1] This established a new ceiling for what AI could achieve in a professional domain, setting the stage for the next frontier in scientific research.

The transition of this AI power to the sciences is already underway, manifesting in a sharp increase in advanced reasoning workloads. Data from an internal OpenAI analysis highlights the speed of this adoption, showing that the average weekly message count on advanced science, technology, engineering, and mathematics (STEM) topics within ChatGPT grew by approximately 47% in the previous year, culminating in nearly 8.4 million weekly messages by early this year.[3][4] This usage spans physics, chemistry, biology, and engineering, underscoring AI's function as a tool for solving open problems, interpreting complex datasets, and accelerating iterative experimental work. Kevin Weil noted that the pace of adoption and the quality of work emerging from this engagement indicate that science is entering a new acceleration phase.[3][4] Crucially, the newest generation of large language models, specifically GPT-5.2, is reported to have “moved from a competitive level towards mathematical research,” achieving a new baseline of expertise across diverse scientific disciplines.[3][4]

Concrete breakthroughs illustrate the potential for profound impact across fields. In mathematics and theoretical computer science, where structured reasoning is paramount, the AI has shown an ability to generate novel proofs. In one notable case, an AI system utilizing a custom math scaffolding around GPT-5 successfully produced short, elegant proofs for two inequalities in graph theory, including one that had previously been an unproven conjecture.[5] In the realm of experimental science, the gains have been equally significant; in a complex biological study, scientists had spent months trying to understand a puzzling change in human immune cells, but GPT-5 was able to analyze unpublished data and identify the likely mechanistic hypothesis within minutes, suggesting an experiment that was later confirmed in the lab.[5] These examples demonstrate AI’s capacity to bypass cognitive and logistical bottlenecks inherent in modern research—from literature review and synthesis to complex computation and hypothesis generation—thereby shortening the cycle from initial idea to validated result.[3]

This acceleration phase holds immense implications for global challenges, particularly in medicine. With noncommunicable diseases still accounting for a vast majority of global deaths and the typical drug development cycle spanning 10 to 15 years, reducing the timeline for discovery can directly save and improve lives.[3] Weil’s ultimate vision is ambitious: to compress the next 25 years of scientific discovery into just five years by using AI as a new type of scientific instrument.[6] However, this optimistic outlook is tempered by a call for increased humility, acknowledging that while AI can shorten research workflows and help researchers move faster toward correct results, it does not yet run projects or solve scientific problems autonomously.[5] The most common current uses of the AI by scientists remain in writing and communication, with analysis and calculations making up the smallest proportion, indicating that the technology is still early in its journey toward becoming a fully integrated, high-throughput partner for thought and computation.[4] As AI models continue to evolve in their ability to handle highly technical derivations and function as reliable workflow orchestrators, the promised scientific revolution of 2026 is expected to cement AI’s role as the most powerful engine for human discovery yet created.[3]