AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

OpenAI’s latest internal report says researchers are using coding agents more intensively, with average agent runtime rising from 0.5 to 5.7 eight-hour workdays per researcher per day. The company describes the findings as a partial measure of AI’s contribution to research and says they should not be read as an intention to pursue full recursive self-improvement.

OpenAI has published new internal measurements showing that coding-agent use in its research organization has risen sharply since its September report, including an increase in average agent runtime from 0.5 to 5.7 eight-hour workdays per researcher per day. The update offers a partial account of how AI tools are being used in the development of future AI systems, an area the company says labs should measure and publicly report.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

OpenAI says the median researcher used agentic AI tools worth about $2,400 per day at API list prices, while use by researchers at the 90th percentile was about $23,000 per day. The company compared those figures with $1,500 and $11,000, respectively, in September, describing the increases as 62% and 105%. These spending estimates capture multiple changes, including activity levels, model choice and changes to models and agent software. They are not a direct measure of research output.

The report also uses “agent workdays” to estimate how long agents run: each eight hours of agent use counts as one workday. OpenAI says the overall average reached 5.7 workdays per researcher per calendar day, up from 0.5 in its earlier analysis. On pretraining and posttraining teams, the average was about eight; on scaling and infrastructure teams, it was about four. The company says this measure generally covers interactive use and may miss some automated activity.

By mid-October, more than two-thirds of the researchers included in the analysis had a daily peak of at least four agents working concurrently, according to OpenAI. More than 10% averaged at least 16 agents at their daily peak. The report also compares agent spending with non-equity researcher compensation: as of September 2026, OpenAI estimated the API-equivalent value of agent use at about twice that compensation. It says internal deployment accounts for 2% of total research compute.

At a glance
reportWhen: Published after the September 2026 repo…
The developmentOpenAI has published an update based on internal measurements of coding-agent use and other indicators of how AI is changing its research process.
AI Research · Briefing

Research Acceleration The Jagged Frontier

OpenAI reports heavier use of coding agents in its research organization. The figures show activity and resource use, but do not establish faster or better research.

5.7
agent workdays per researcher per day, up from 0.5
$2,400
median daily agent use at API list prices
$23,000
daily agent use at the 90th percentile
2%
of total research compute from internal AI deployment

Agent use is rising

Eight hours of agent use counts as one agent workday.

September
  • 0.5 average agent workdays per researcher per day
  • $1,500 median daily agent use
  • $11,000 daily use at the 90th percentile
Latest report
  • 5.7 average agent workdays per researcher per day
  • $2,400 median daily agent use
  • $23,000 daily use at the 90th percentile

Use differs across teams

Average agent runtime varies by research area.

Pretraining and posttrainingAbout 8
Scaling and infrastructureAbout 4

More agents work concurrently

By mid-October, many researchers had several agents active at their daily peak.

Concurrency

More than two-thirds

Had a daily peak of at least four agents working at once.

High peaks

More than 10%

Averaged at least 16 agents at their daily peak.

What the figures leave open

The measures capture inputs to research, not the effects on its outcomes.

1
Activity is not output

Runtime and spending do not show whether agents produced useful work or improved research pace, quality, or success.

2
Costs are partial

API list prices reflect changes in usage, model prices, and agent behavior; the compensation comparison excludes equity.

3
Coverage is incomplete

The workday measure may miss automated activity, and the report describes its account as partial.

ContextOpenAI says the report is not an indication that it intends to pursue full recursive self-improvement.
The takeaway
Coding agents are becoming a substantial part of researchers’ workflows, but comparable measures of research outcomes are needed to show what that use achieves.

How Agent Use Is Changing Research

The measurements describe a shift in the tools and resources researchers use to develop AI systems. If coding agents take on more work or operate for longer periods, they could affect how quickly research teams test ideas and build systems. OpenAI’s figures show substantial growth in use inside its own research organization, but they do not establish how much faster research is progressing or how much of that progress is attributable to AI.

The report matters beyond the company because it makes a case for tracking indicators of recursive self-improvement, the process by which AI could help accelerate the development of later AI systems. OpenAI says such measures could inform discussion of capabilities, risks and safeguards. It also says the update is not an indication that the company intends to pursue full recursive self-improvement.

Some of the figures may be easy to misread. Agent spending is valued using API list prices, while the compensation comparison excludes equity, which OpenAI says generally makes up 70% to 80% of researchers’ target compensation. The company also notes that internal AI deployment is a small share of total research compute. Together, these limits make the figures useful as indicators of activity and allocation, but not a complete account of research costs, productivity or impact.

Products Worth Considering

Bose QuietComfort Wireless Noise-Cancelling Headphones, Twilight Blue

Bose QuietComfort Wireless Noise-Cancelling Headphones, Twilight Blue

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

As an affiliate, we earn on qualifying purchases.

From September Baseline to October

This update follows an OpenAI report published in September on research acceleration. The company says the new report adds quantitative and qualitative measures of AI’s contribution across research areas and of continued human involvement. It organizes the subject around inputs to research, the process of converting those inputs into model capabilities, and resulting changes in capabilities.

The new figures focus mainly on coding agents as one input. OpenAI says research also depends on data, human labor and compute, but the report’s agent-use measures do not cover all of those factors. Its spending comparison tracks monthly API-equivalent agent use against monthly non-equity compensation; the stated labor-cost measure includes salary, overtime, bonuses, benefits and employer taxes, but excludes equity.

OpenAI frames the work as part of a broader push for shared measurement practices. The company says other organizations have also contributed to efforts to measure AI research and development acceleration. Its stated rationale is that clearer reporting could help the public consider how oversight and safeguards should change as AI systems gain capability.

What the Measurements Leave Open

The report does not establish how much agent use has changed the pace, quality or success rate of OpenAI’s research. It reports spending and runtime indicators, but those figures do not by themselves show whether agents produced useful work, displaced researcher effort or improved model capabilities. The source material also describes planned analysis of researchers’ contributions and time allocation, but does not provide the results of that analysis here.

The scope of the measurements is limited. The agent-workday figure generally includes interactive use and may omit automations, while list-price spending reflects changes in usage, model prices and agent behavior. The compensation ratio excludes equity, and the company describes its wider account as partial. The report does not provide enough detail in the supplied material to assess how representative the researchers included are or how the measures were calculated across teams.

More Reporting on Research Inputs

OpenAI says its report is intended to contribute to shared measurement norms for AI research and development. The company presents inputs, research processes and capability changes as related areas for assessment. Further reporting would help readers understand how agent use connects to research outputs and what role human researchers continue to play.

The report does not give a schedule for another update or specify a common measurement standard adopted across AI labs. OpenAI’s figures therefore provide a dated snapshot: spending is reported as of September 2026, while the concurrency analysis reaches mid-October. Whether other organizations will publish comparable measures, and whether those measures can support meaningful comparisons, remains to be seen.

Where I land

I read this as a useful disclosure about how heavily one AI lab is beginning to rely on coding agents, and as a case for making these measurements more visible. The rise in runtime and concurrency deserves attention because it shows that agent use is becoming a significant part of researchers’ daily workflows. That is a meaningful development even before it can be linked to better or faster research outcomes.

The strongest counterargument is that the report measures activity more clearly than effect. Spending at list prices and accumulated runtime can rise without a matching increase in useful research, and the compensation comparison excludes a large share of target pay. I would change my assessment if comparable reports showed, with clear methods and output measures, that greater agent use reliably speeds up research while preserving human oversight and producing capabilities that can be evaluated independently.

Key Questions

What did OpenAI report?

OpenAI reported increases in coding-agent use among researchers, including higher estimated daily spending, longer agent runtime and more concurrent agents in the measurements it describes.

What is an agent workday?

OpenAI defines one agent workday as eight hours of agent use. It says the average reached 5.7 workdays per researcher per calendar day, while noting that the measure generally covers interactive use and may miss automations.

Does the report show that AI research has accelerated?

It documents increased use of AI tools and discusses research acceleration, but the figures provided do not establish how much faster research is progressing or what improvements in model capabilities resulted.

Is OpenAI pursuing recursive self-improvement?

OpenAI says the report supports measuring indicators of recursive self-improvement, but says its reporting should not be read as an intention to pursue full recursive self-improvement.

Source: OpenAI

You May Also Like

The Deploy Button Became the Bottleneck — and Cloudflare Just Bought the Build Step

AIThis post was created with the assistance of artificial intelligence (AI).Something has…

Minerva. The opposite path.

AIThis post was created with the assistance of artificial intelligence (AI).Italy spent…

Ranking in ChatGPT: The New Playbook for AI Search Visibility (and How to Use ZimmWriter)

AIThis post was created with the assistance of artificial intelligence (AI).“Ranking in…

Claude for Word Is Not a Feature. It’s a Market Signal.

AIThis post was created with the assistance of artificial intelligence (AI).Anthropic just…