A story we follow
OpenAI Reports Rising Internal Research-Agent Use
Tracks OpenAI's September research-acceleration report, its internal agent-use metrics and interpretations of research productivity.
A story page follows one specific event across podcast discussions, with an overview and a timeline of what changed. It updates when new episodes discuss the event, and you can get those updates by email or push. How our formats work
By email or push, when we publish an update to this story. It does not subscribe you to other stories.
Overview
OpenAI's research-acceleration report describes rising internal coding-agent activity and improving task outcomes. Moonshots read its 3.1 figure as completed research work exceeding that of human interns. The report instead measures total agent runtime per human workday, not equivalent scientific output, and notes human interventions and measurement limits. OpenAI researcher Noam Brown cited the report on the Dwarkesh Podcast. From memory, he said the top 1 per cent of researchers were spending about 7,000 to 8,000 dollars a day on Codex as of early August. Background notes to the episode say the report puts spending above 7,000 dollars at the 90th percentile. Brown said acceleration is hard to measure because of attribution and uneven task speed-ups. His rough estimate was that research could go about three times faster, within a range from 50 per cent to an unlikely tenfold. The link between higher usage and research progress remains unresolved.
Recent coverage
Changes to our coverage published in the past seven days. These are site update dates; podcast discussion dates and sources are in each version.
No changes to this account were published in the past seven days.
What changed
Dates show when each podcast discussion was published.
-
New information
OpenAI researcher Noam Brown cited the internal-acceleration report on the Dwarkesh Podcast. He recalled that as of early August, the top 1 per cent of researchers were spending about 7,000 to 8,000 dollars a day on Codex for internal use, and that this will keep rising. Background notes to the episode say the report attaches spending above 7,000 dollars to the 90th percentile. Brown said acceleration is hard to measure for three reasons: it is unclear how to split credit between humans and AI, some tasks speed up far more than others, and baseline comparisons differ. He said he is confident research is faster than a year ago. His guess was about three times faster, with a range from 50 per cent to an unlikely tenfold.
-
New information
The hosts present OpenAI's internal usage figures as evidence that agents have surpassed research interns and may enable recursive improvement. Their output-equivalence interpretation goes beyond the runtime metric in the report.
Podcast discussions
- Dwarkesh Podcast Noam Brown – Agent swarms, alignment, & recursive self-improvementSep 17, 2026
- Moonshots with Peter Diamandis Why Jensen Huang Believes We’ve Reached AGI and Inside OpenAI’s German Website Hijack | #287Sep 9, 2026
Sources
- 01
- 02
- 03
Our coverage
Huang declares AGI; OpenAI's 3.1 figure answers a narrower question
Nvidia's Jensen Huang declared AGI achieved on September 6 as he announced more GPU capacity, and the Moonshots panel split on what the label means. OpenAI's figure of 3.1 agent research days per human day measures runtime, not finished work.
Version history
-
Sep 29, 2026 · Version 2
OpenAI researcher Noam Brown cited the internal-acceleration report on the Dwarkesh Podcast. He recalled that as of early August, the top 1 per cent of researchers were spending about 7,000 to 8,000 dollars a day on Codex for internal use, and that this will keep rising. Background notes to the episode say the report attaches spending above 7,000 dollars to the 90th percentile. Brown said acceleration is hard to measure for three reasons: it is unclear how to split credit between humans and AI, some tasks speed up far more than others, and baseline comparisons differ. He said he is confident research is faster than a year ago. His guess was about three times faster, with a range from 50 per cent to an unlikely tenfold.