Research acceleration: an inside view from OpenAI
Research acceleration: The view inside OpenAI
OpenAI says it achieved an automated research intern in September of this year, as announced as a goal last fall, meaning it can complete well-defined research tasks under human guidance that would take a skilled researcher days, and says it is progressing toward an automated AI researcher by March 2028. Its disclosed internal data shows that by mid-August, the median research user used more than $600 of API-priced inference per day, the 90th percentile user used more than $7,000 of tokens per day, and the entire research organization had 3.1 agent workdays for every 1 human workday; August had the highest number of experiments per active experimenter since tracking began in January 2025. OpenAI says agents still require substantial human guidance, and in 4-8 hour tasks, more than half of successful cases involved at least one intervention.
OpenAI discloses internal data showing how coding agents accelerate research workflows and how safety restrictions affect compute allocation.