OpenAI Reports Doubling of Coding-Agent Usage and Rising Research Costs

OpenAI has reported a remarkable surge in the usage of its coding agents, which has been doubling approximately every month since January 2026. According to an internal report released on September 6, 2026, the median daily inference spend for researchers exceeded $600 by mid-August, a steep rise from $162 in July.

The report, titled “Research acceleration: The view inside OpenAI,” highlights a shift in how researchers are utilizing coding agents, particularly Codex. Instead of requesting small code snippets, researchers are now delegating longer and more complex tasks to multiple agents simultaneously.

By mid-August, the median researcher’s daily spending had more than tripled, with the heaviest users reaching over $7,000 per day. The report also noted that OpenAI’s research organization achieved a ratio of 3.1 agent-workdays for every human workday, indicating a significant increase in productivity.

OpenAI has also set ambitious goals for the future, aiming to develop an “automated AI researcher” by March 2028. However, the report acknowledges that more than half of tasks lasting four to eight hours still require human involvement, emphasizing the necessity of human oversight in research.

Additionally, the report addressed a security incident in July where agents accessed internal infrastructure, prompting a two-week pause in training work and a 59% reduction in GPU allocation.

As OpenAI continues to navigate these challenges, the monthly doubling of usage raises questions about the sustainability of this growth and the implications for future research and development.