The term "PhD-level AI" has emerged in the AI industry, with a report from The Information suggesting that OpenAI is set to launch specialized AI "agents. " These include a $20, 000 monthly tier aimed at enhancing "PhD-level research, " alongside a $2, 000 agent for high-income knowledge workers and a $10, 000 software developer agent. Although OpenAI has not confirmed these pricing plans, they have previously discussed AI models with abilities akin to doctoral expertise. "PhD-level AI" refers to models capable of carrying out advanced research, programming complex code, and analyzing large datasets. OpenAI's assertions are partially backed by benchmark test performances; for example, their o1 models demonstrated results similar to human PhD students in science, coding, and math tasks. Their Deep Research tool also performed notably on extensive evaluations. OpenAI's recent innovations include the o3 and o3-mini models, announced in December, which utilize a method called "private chain of thought, " allowing the model to engage in simulated reasoning and thus tackle complex problems more effectively over time. The o3 model achieved impressive scores on various benchmarks, including a record on the ARC-AGI visual reasoning test, and an outstanding score on the American Invitational Mathematics Exam. Potential real-world applications for such advanced AI could encompass medical research data analysis, climate modeling, and automating routine research tasks.

The high costs associated with these AI agents reflect OpenAI's belief in their significant business value, particularly with investor commitment from SoftBank. OpenAI's pricing contrasts sharply with more affordable AI solutions like ChatGPT Plus and Claude Pro, raising questions about the tangible returns on such high-priced offerings. Despite their benchmark successes, concerns remain about these models' tendency to produce misleading information, particularly in critical research scenarios. Critics highlight that hiring actual PhD students is more cost-effective, pointing out that the label “PhD-level AI” serves more as marketing than a reflection of true intellectual capabilities. While these AI systems are impressive, the complexities of creative thinking and original research still present challenges. However, they remain tireless and are anticipated to improve and decrease in cost over time.