OpenAI's CTO: GPT-5 Expected to Debut by Late 2025 or Early 2026, Reaching 'PhD-Level' Intelligence

On June 22, the Dartmouth School of Engineering released an engaging interview with Mira Murati, Chief Technology Officer of OpenAI. Murati compared the evolution from GPT-4 to GPT-5 to the academic journey from high school to doctoral studies.

“Models like GPT-3 represent a toddler's intelligence, while GPT-4 mirrors the smarts of a bright high school student. We anticipate that in the coming years, we will achieve doctoral-level intelligence in specific tasks,” she explained. Murati emphasized that this "doctorate-level" intelligence will only pertain to certain areas. "These systems already demonstrate human-level performance in some tasks, but they certainly fall short in many others."

When discussing the timeline for GPT-5, Murati indicated that a release is expected within the next year and a half, potentially launching by late 2025 or early 2026. This estimation is consistent with insights from other industry leaders. Microsoft CTO Kevin Scott noted that the next generation of AI systems will feature improved memory and reasoning skills, even achieving the capability to pass doctoral exams. Alibaba Chairman Joe Tsai also highlighted the rapid advancements in large language models, remarking, “In just three to four years, we've reached human-equivalent knowledge and certain mathematical skills, with some models even exhibiting doctoral-level capabilities. This rapid progress is both astonishing and intimidating.”

This discussion underscores the remarkable pace of AI advancements and stimulates intriguing questions about the future trajectories of intelligent systems.