OpenAI official logo (public domain, Wikimedia Commons) — CryptoBriefing brand treatment
An internal OpenAI report says agents now log 3.1 workdays for every human workday, and the compute bill is climbing just as fast
OpenAI’s use of coding agents has been doubling roughly every month since January 2026, according to Epoch AI. The bill is growing just as fast.
An internal OpenAI report released on September 6, 2026, says the median researcher’s daily inference spend passed $600 by mid-August. In July, that figure was $162.
What the report actually says #
The report is titled “Research acceleration: The view inside OpenAI.” It describes how coding agents, Codex in particular, have changed the way the company’s researchers work.
The basic shift is about delegation. Researchers are no longer asking agents for small code snippets. They are handing off longer, more complex tasks and running several agents at the same time.
The spending numbers show how far that has gone. The median researcher crossed $600 per day in inference costs by mid-August, more than triple the $162 recorded in July. At the 90th percentile, the heaviest users topped $7,000 per day.
The headline labor metric is just as striking. By mid-August, OpenAI’s research organization logged 3.1 agent-workdays for every human workday, a ratio the report says it reached after June 2026.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Output is up, and so are the guardrails #
The report links the agent surge to measurable gains. Experiments per active researcher hit an all-time high in August 2026. Experiment throughput and code changes also grew faster than activity elsewhere at OpenAI.
OpenAI also says it hit a goal it set for itself: building an “automated research intern” by September 2026. The next target is more ambitious. The company plans an “automated AI researcher” by March 2028.
According to the report, more than half of tasks lasting four to eight hours still need a human involved. People still decide what matters, which experiments deserve priority and when a result can be trusted.
Then there is July. The report describes a security incident in which agents gained access to internal infrastructure. OpenAI responded with a two-week on training work, including certain reinforcement learning activities. The report links the episode to a 59% reduction in GPU allocation.
Why the doubling curve matters #
Epoch AI’s observation that usage doubles about monthly is the figure that frames everything else. The jump from $162 to more than $600 in median daily spend shows that pressure is already arriving.
A few things are worth tracking. First, whether the monthly doubling continues, slows or runs into compute limits. Second, how OpenAI handles agent security after July, since a that cut GPU allocation by 59% is a costly way to learn a lesson. Third, whether the share of four-to-eight-hour tasks that need human oversight starts to drop. That metric, more than any spending figure, will show whether the March 2028 “automated AI researcher” target is a roadmap or an aspiration.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our