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

🔍 Read the full analysis: How AI Is Accelerating Research At The Jagged Frontier on ThorstenMeyerAI.com

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.

TL;DR

OpenAI’s latest internal report says coding-agent use in its research organization has increased sharply, with average runtime rising from 0.5 to 5.7 eight-hour workdays per researcher per calendar day. The company calls these figures partial indicators of AI use, not proof that research is progressing faster or that it plans to pursue full recursive self-improvement.

OpenAI says researchers in its organization are using coding agents for longer periods than in its September analysis, reporting an average of 5.7 eight-hour agent workdays per researcher per calendar day, up from 0.5. The figures offer a partial view of how AI tools are used to develop AI systems, but do not show whether the tools have made research faster or more successful, as discussed in the original analysis.

The company’s report also estimates that the median researcher used agentic AI tools valued at about $2,400 per day at API list prices, compared with $1,500 in September. For researchers at the 90th percentile, the estimate was about $23,000 per day, up from $11,000. OpenAI characterizes those changes as increases of 62% and 105%, respectively.

OpenAI says the spending estimates combine several factors, including how much agents are used, which models researchers select, and changes to models and agent software. They are not a direct measure of useful work or research output. The company defines an “agent workday” as eight hours of agent runtime; the overall figure is an average per researcher per calendar day, not a claim that a person worked that many hours.

Use varied across research teams: OpenAI reports an average of about eight agent workdays on pretraining and posttraining teams and about four on scaling and infrastructure teams. By mid-October, more than two-thirds of researchers covered by the analysis had a daily peak of at least four agents working concurrently, while more than 10% averaged at least 16 agents at their daily peak.

At a glance
reportWhen: The update describes measurements throu…
The developmentOpenAI has published new internal measurements showing increased coding-agent runtime and spending among researchers since its September analysis.
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.

What Higher Agent Use Can Show

The figures document a change in how one major AI lab allocates tools and computing resources to research. Longer agent runtimes and more concurrent agents could let teams delegate coding tasks or run more work in parallel. That possibility makes the measurements relevant to questions about how AI development may change as AI systems become part of the process of building later systems.

But the report does not establish that those possibilities have translated into faster progress, better research, or improved model capabilities. Agent runtime and estimated spending track activity and resource use, not the quality or consequences of the work. They also do not reveal how much human effort agents replace, or how much new work they enable.

OpenAI presents measurement as a way to inform discussion of capabilities, risks and safeguards, and argues that labs should report indicators of AI’s contribution to research. Shared measures could help observers compare developments over time, but only if organizations explain their scope and limitations. The company says its own account is partial and that the update is not an indication it intends to pursue full recursive self-improvement.

Products Worth Considering

As an affiliate, we earn on qualifying purchases.

From September Baseline to October

The new measurements follow an OpenAI report on research acceleration published in September. The update adds quantitative and qualitative measures concerning AI’s contribution across research areas and the continued role of people, but the figures supplied here focus mainly on coding agents as one input to research.

OpenAI says research also depends on human labor, data and compute. Its agent-use measures do not cover all those inputs or provide a complete account of how research turns resources into new model capabilities. The company describes the report as part of a broader effort to develop shared ways of measuring AI’s contribution to research and development.

One comparison sets agent spending against researcher compensation. OpenAI says that, as of September 2026, the API-equivalent value of agent use was about twice non-equity researcher compensation. The compensation measure includes salary, overtime, bonuses, benefits and employer taxes but excludes equity; OpenAI says equity generally accounts for 70% to 80% of researchers’ target compensation. The company also reports that internal AI deployment accounts for 2% of total research compute.

What Runtime Cannot Establish

The report’s activity measures do not establish how much agent use has affected the pace, quality or success rate of research. It remains unclear whether agents produced useful work, displaced tasks previously done by researchers, or helped produce changes in model capabilities. The source material also refers to planned analysis of researchers’ contributions and time allocation but does not give its results.

The measures have other limits. OpenAI says the workday figure generally covers interactive use and may miss some automated activity. API list-price estimates can change with usage, model prices and agent behavior. The compensation ratio excludes equity, and the supplied material does not give enough detail to judge how representative the researchers included are or how calculations were made across teams.

OpenAI’s claim that agent use was worth about twice non-equity compensation is therefore a comparison of specified estimates, not a complete accounting of research costs or a productivity measure. The report also says internal AI deployment represents 2% of total research compute, but that figure alone does not explain how much research work agents performed.

Further Reporting on Research Inputs

OpenAI says the report is intended to contribute to shared measurement practices for AI research and development. The company says other organizations have also contributed to efforts to measure acceleration. The supplied material does not give a date for another report or specify when the planned analysis of researcher contributions and time allocation will be released.

For a clearer picture of impact, further reporting would need to connect agent activity with research outcomes and account for other inputs, including human work, data and compute. Until then, the new figures show that coding agents are being used more intensively within OpenAI’s research organization, while the effects on research progress remain unresolved.

Key Questions

What did OpenAI report about coding-agent use?

OpenAI says average agent runtime reached 5.7 eight-hour workdays per researcher per calendar day, up from 0.5 in its September analysis. The measure estimates time agents ran, not human working hours or research output.

How much did researchers’ estimated agent spending rise?

OpenAI estimates median daily use at about $2,400 at API list prices, up from $1,500 in September. For researchers at the 90th percentile, it reports about $23,000 per day, compared with $11,000. The company says these estimates reflect several factors and are not direct productivity measures.

Does the report show that AI has made OpenAI’s research faster?

No. The report describes agent runtime and estimated spending but does not establish changes in research speed, quality, success rates or model capabilities.

Does the report mean OpenAI plans full recursive self-improvement?

OpenAI says the update is not an indication that it intends to pursue full recursive self-improvement. The report argues for measuring AI’s contribution to research as systems are used to help develop later systems.

Primary source: OpenAI · via ThorstenMeyerAI.com

You May Also Like

An AI coding agent, used to write code, needs to reduce your maintenance costs

An AI coding agent’s effectiveness depends on lowering long-term maintenance costs, not just increasing code output, to ensure sustained productivity gains.

How DaisyRose-27B Advances Post-Training Efficiency

RadixArk says its DaisyRose-27B model improved Terminal-Bench 4.1 pass@1 to 12%. Independent checks and training details remain outstanding.

YouTube to automatically label AI-generated videos

YouTube will now automatically apply labels to videos with significant photorealistic AI use, enhancing transparency for viewers starting in 2024.

Why Global South AI Innovation Deserves More Attention

Unlock the potential of Global South AI innovation to transform local challenges into global opportunities, revealing why this movement deserves more attention.