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

🔍 Read the full analysis: Research Acceleration: 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 average coding-agent runtime in its research organization rose from 0.5 to 5.7 eight-hour workdays per researcher per calendar day since its September report. The company presents agent use and spending as partial indicators, not proof that AI has accelerated research or a plan to pursue full recursive self-improvement.

OpenAI’s latest internal report says researchers in its organization used coding agents for an average of 5.7 eight-hour workdays per researcher per calendar day, up from 0.5 in the company’s September report. The figures offer a partial measure of AI use in research, but do not establish that the work is progressing faster or show how much of any research result comes from agents, as discussed in the original analysis.

OpenAI also estimated the daily value of agent use at API list prices. The median researcher’s use was about $2,400 per day, compared with $1,500 in September; for researchers at the 90th percentile, the estimate was about $23,000, up from $11,000. The company described those changes as increases of 62% and 105%, respectively. It says spending can reflect multiple factors, including usage levels, model selection, and changes to models and agent software, rather than research output alone.

Agent runtime varied across teams. OpenAI reported an average of about eight agent workdays per researcher per calendar day on pretraining and posttraining teams, and about four on scaling and infrastructure teams. The measure counts each eight hours of agent use as one workday. OpenAI says it generally covers interactive use and may miss some automated activity. For related coverage, see the view inside OpenAI.

Concurrency was also common among the researchers included in the analysis: by mid-October, more than two-thirds had a daily peak of at least four agents working at the same time, according to the company. More than 10% averaged at least 16 agents at their daily peak. OpenAI additionally said internal deployment represented 2% of its total research compute.

At a glance
reportWhen: Updated report published after the Sept…
The developmentOpenAI published updated internal measurements showing increased coding-agent runtime and spending among researchers since its September report.
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 Agent Use Can—and Cannot—Show

The measurements matter because they document a marked change in how one major AI developer allocates tools and resources inside its research organization. Longer agent runtimes and higher estimated spending could affect how researchers test ideas and build systems. But those are activity and resource indicators, not a measure of the pace, quality, or success rate of research.

OpenAI presents the report as part of a broader case for shared ways to measure whether AI systems contribute to the development of later systems, sometimes called recursive self-improvement. Better reporting could help inform public discussion of AI capabilities, risks, and safeguards. The company says this update should not be read as an intention to pursue full recursive self-improvement.

The financial comparisons have limits. The agent-use figures rely on API list prices, rather than a direct accounting of the company’s costs. The comparison with researcher compensation excludes equity, which OpenAI says generally represents 70% to 80% of researchers’ target compensation. The company also describes the overall measurements as partial, so they cannot on their own show the full cost or impact of AI use in research.

Products Worth Considering

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray

Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite

Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite

As an affiliate, we earn on qualifying purchases.

From September Baseline to October

The new report follows an OpenAI analysis published in September on research acceleration. The company says the update adds quantitative and qualitative measures of AI’s contribution across research areas, alongside consideration of continued human involvement. The figures supplied here focus mainly on coding agents as one input to research.

OpenAI’s framework describes research as involving inputs, the process of turning those inputs into model capabilities, and changes in capabilities. Agent use is only one part of that picture: the company also identifies data, human labor, and compute as inputs. The runtime and spending measures do not account for all of them, nor do they independently track the results of research.

For its compensation comparison, OpenAI says it compared monthly API-equivalent agent use with monthly non-equity compensation. Its labor-cost measure includes salary, overtime, bonuses, benefits, and employer taxes, but excludes equity. The company says other organizations have also contributed to efforts to measure how AI may accelerate research and development.

What the Figures Leave Unanswered

The report does not establish how much agent use changed research speed, quality, or success, or how much useful work agents produced. It also does not show whether agents displaced researcher effort, complemented it, or had different effects across projects. The supplied material mentions planned analysis of researchers’ contributions and time allocation but does not provide those results.

Several details needed to assess the estimates are not available in the supplied material, including how representative the researchers in the analysis were and how the measures were applied across teams. Runtime may omit some automated activity; list-price spending can change with usage, models, and software; and the compensation comparison excludes equity. These limits mean the reported figures should be read as partial indicators, not a complete account of research costs or impact.

Toward Shared Research Measures

OpenAI says it wants the report to contribute to shared measurement norms for AI research and development acceleration. The next step described in the material is broader reporting and analysis of how AI contributes across research inputs, processes, and resulting capabilities, including the role people continue to play.

No date for a further report or a specific next measurement was provided. Until more outcome-focused evidence is available, the runtime and spending estimates show how extensively researchers are using coding agents, but not whether that use has produced faster or better research.

Key Questions

What changed in OpenAI’s latest report?

OpenAI reported that average coding-agent runtime rose from 0.5 to 5.7 eight-hour workdays per researcher per calendar day compared with its September report. It also reported higher estimated agent spending at API list prices.

Does the report prove that AI is speeding up research?

No. The figures measure agent runtime and estimated spending, but do not establish changes in research speed, quality, or success, or show how much agents contributed to results.

What does an agent workday mean?

OpenAI counts eight hours of agent use as one workday. The company says the measure generally captures interactive use and may not include some automated activity.

Is OpenAI saying it plans to pursue recursive self-improvement?

No. OpenAI says the report’s measurements are not an indication that it intends to pursue full recursive self-improvement. It presents measurement as relevant to discussion of capabilities, risks, and safeguards.

Why are the spending estimates not a complete measure of research cost?

The estimates use API list prices and can reflect changes in usage, models, and agent software. The compensation comparison excludes equity, and the company says internal deployment accounts for 2% of total research compute.

Primary source: OpenAI · via ThorstenMeyerAI.com

You May Also Like

Interviewing Engineers In The AI Era: Lessons From A Year Of Rebuilding

An in-depth analysis of how companies are redesigning engineer hiring amid rapid AI advancements, based on a year of industry interviews and experiences.

The Last AI Built By Humans: Toward Genuine Recursive Self-Improvement

Researchers announce the development of the final AI built by humans capable of genuine recursive self-improvement, marking a milestone in AI evolution.

Europe’s $14 Billion AI Bet: The Mistral Initiative For Sovereignty And Innovation

Europe invests $14 billion in Mistral to build a sovereign AI model, emphasizing data residency, infrastructure independence, and industrial policy.

Humans Missed 1 In 3 Threats Approving AI Agent Commands Across 40K Game Runs

Study finds humans failed to identify 33% of threats in 40,000 AI-driven game simulations, raising concerns over oversight in AI safety.