OpenAI says agents now log 3.1 workdays for each human research day
OpenAI published an internal snapshot of coding agent use across its research group. By mid-August, OpenAI says total agent runtime equaled 3.1 eight-hour workdays for every human workday. The median researcher was using more than six hundred dollars of inference each day at API prices. A user at the ninetieth percentile was using more than seven thousand dollars. OpenAI says researchers are writing code and running experiments faster. It now calls the system an automated research intern. The label covers well-defined jobs that can take a skilled researcher several days.
Verified 2:45 AM PDT · 1 original sources
OpenAI bases these claims on its own internal data. Agent runtime is not research output, and OpenAI says its measurements are preliminary. Available compute also grew. High-level planning remained a small share of agent output. More than half of successful four-to-eight-hour tasks still needed at least one human intervention. The company targets an automated AI researcher by March 2028, but that is a goal, not an observed result.
The useful measure is accepted research work per dollar and per human review hour. Watch for outside replication of the task taxonomy, measured experiment quality and the rate at which agent work survives review. OpenAI should also report whether more agent runtime shortens model development, or whether it moves the bottleneck to compute, safety review and research judgment.
Audit the story
Original sources
Company claims remain company claims. Follow the reporting and judge the evidence directly.
Continue the edition