Epoch AI has applied a new growth model to earlier OpenAI researcher usage data, estimating a recent median doubling time of 34 days. The series runs from January to mid-August and values activity at published API prices. These values represent neither actual bills nor internal computing costs. The analysis includes zero-use researchers but excludes automated background Codex exec work.
Epoch AI, an organisation studying artificial intelligence, published a new analysis of OpenAI researchers’ coding-agent use. Researchers Yafah Edelman and Luke Emberson estimate that recent median usage grew by about 1.8 times a month. Their model puts the doubling time at 34 days, with a 90 per cent confidence interval from 29 to 42 days. The October development is the new growth estimate applied to data OpenAI released in September.[1]
The underlying series follows daily use per researcher from January to mid-August, smoothed into weekly averages. Activity is valued using published application programming interface, or API, prices from early September. Median daily usage moves from under 1 dollar to 601.25 dollars over the period. These are price-equivalent usage values, rather than the amounts OpenAI paid or its internal infrastructure costs. The upper-use measure at the 90 per cent threshold reaches 7,047.13 dollars daily.[1]
Epoch AI compares a single exponential curve with models that divide growth into periods around a breakpoint. Its selected upper-use model implies roughly 2.2-fold monthly growth and a 27-day doubling time. The 90 per cent confidence interval for that time is 26 to 30.5 days. The dataset includes people with zero use and research-organisation functions such as management and infrastructure. It measures interactive activity, excluding automated Codex exec jobs, rather than following only an unchanged panel of researchers.[1]
