Coding-agent use at OpenAI is doubling roughly every month
B · Useful, limited Short Epoch AI data insight that fits trends to usage data self-reported by OpenAI; the method is transparent, but the figures come from one company and are priced at list rates.
Summary based on the full text·Original: about 6 minutes to read
In short
A data insight published by Epoch AI on 5 October 2026, written by Yafah Edelman and Luke Emberson. It extracts the data behind charts in OpenAI's September 2026 post on how its researchers use coding agents and fits growth trends to them. The median researcher's daily coding-agent use, valued at API list prices, went from under $1 in January 2026 to $601 by mid-August. Recent growth implies a doubling roughly every month for both typical and heavy users.
Key numbers
- $601.25median OpenAI researcher's daily coding-agent use at API prices, week ending 15 Aug 2026
- $7,047.1390th-percentile researcher's daily use in the same week
- 34 / 27 daysrecent doubling time for median and 90th-percentile users
- about $2 millionannual run rate at API prices for a 90th-percentile researcher (260 working days)
As reported by the source.
English·Web page·opens epoch.ai
What it reports
- Median researcher daily spend: $0.17 (week ending 18 January), $10.45 (8 March), $136.68 (31 May), $601.25 (15 August).
- 90th-percentile researcher: $1.95 (4 January), $112.15 (8 March), $972.41 (31 May), $7,047.13 (15 August).
- Preferred model (exponential with one breakpoint): after the breakpoint, growth is 1.8x per month for the median (doubling in 34 days, 90% CI 29-42) and 2.2x for the 90th percentile (27 days, CI 26-30).
- Before the breakpoint, growth was 6.3x per month (median) and 15x per month (90th percentile).
- A simple single-trend fit gives 2.7x and 2.4x per month, doubling in 21 and 24 days.
- Final weekly values annualise to about $160,000 per year for the median researcher and about $2 million for the 90th percentile, comparable to or above the cost of employing a researcher.
- If recent growth continued another year, run rates would reach about $185 million (median) and $2.3 billion (90th percentile); Epoch calls this probably unsustainable.
How they got it
Epoch extracted OpenAI's weekly series, where each point averages daily percentiles across all research-organisation staff, including those with zero use. OpenAI priced usage at retail API rates as of 3 September 2026, mapped internal models to their nearest released model, and excluded programmatic automations such as Codex exec. Epoch fit linear and piecewise-linear trends to log values, chose models by Bayesian information criterion, and bootstrapped confidence intervals.
Limits
- The data are self-reported by OpenAI and cannot be verified; list prices say nothing about OpenAI's actual inference cost.
- Percentiles are cross-sectional, so the population changes week to week; early median growth mixes adoption with intensity of use.
- The 90th-percentile breakpoint date shifts between 25 January and 8 February depending on settings, so early growth rates are less robust.
Why it matters for companies
This is one of the few public figures on what heavy agent use costs at market prices: for top users the compute bill can match a salary. Companies budgeting for coding or knowledge agents should expect consumption to rise steeply once people adopt them, with a small group of power users driving most of the spend. Finance and IT will need per-user usage tracking and a view on which work justifies that cost. OpenAI runs its own models; outside buyers pay the full list price.
Summary based on the full text. Read the original for the full argument and data.Read the original on epoch.ai