OpenAI's AI Agents Achieve 3:1 Research Productivity Over Human Staff

OpenAI has announced a major advancement in its AI capabilities, revealing that its AI agents are now performing research tasks at a ratio of 3.1 to 1 compared to human researchers. This milestone, achieved by mid-August 2026, indicates a substantial shift in the reliance on AI for complex research tasks that previously required skilled human input.

The company had set a goal in fall 2025 to develop an “automated research intern,” which has now been realized. As of September 2026, OpenAI reported record levels of experimental activity, with the highest number of experiments conducted per active researcher since tracking began in January 2025.

Despite the impressive productivity of AI agents, the cost of running these systems is significant, with median researchers incurring daily inference costs exceeding $600 and the top 10% spending over $7,000 daily. The tasks managed by AI agents have expanded to include troubleshooting, monitoring experimental runs, and executing sophisticated research projects, although high-level planning still requires human oversight.

OpenAI's next milestone is to achieve a fully automated AI researcher by March 2028, emphasizing the importance of safety and transparency in its development process. The company also acknowledged the concept of recursive self-improvement, highlighting its commitment to responsible AI advancement.