On October 5, 2026, Epoch AI published a data insight by Yafah Edelman and Luke Emberson fitting trends to data OpenAI released in a post on research acceleration inside the company. Valued at OpenAI’s API list prices, the median OpenAI researcher’s daily coding-agent usage rose from under $1 in January 2026 to $601 by mid-August. The 90th percentile researcher was spending over $7,000 a day by mid-August.
Epoch’s preferred fit, an exponential trend with one breakpoint, gives recent growth of 1.8 times per month for the median researcher and 2.2 times for the 90th percentile, doubling times of 34 and 27 days. Both series grew several thousandfold from their first positive January values. At the final weekly values, that is about $160,000 a year at API prices for the median researcher and about $2 million for the 90th percentile, comparable to or more than the cost of employing a researcher. Epoch calls the pace probably unsustainable: another year at that rate would mean annual run rates of about $185 million and $2.3 billion per researcher.
OpenAI counted only interactive product use by members of its research organization, priced at September 3, 2026 retail API rates, and excluded programmatic automation such as Codex exec.
Why it matters: it is a rare quantitative look inside a frontier lab at how fast its own researchers are handing work to agents, which is the mechanism behind forecasts of AI accelerating AI research.
What it does not show: the data are self-reported by OpenAI and cannot be checked independently. API list prices are not OpenAI’s internal cost, so these are not real expenditures, and the percentiles are cross-sectional, so early median growth mixes adoption with intensity of use.