Skip to main content
Live
Main content

OpenAI says coding agents now finish 46% of its 2-4 hour research tasks

The success rate doubled from 23% in March, and median researcher agent use hit about $2,400 a day at API list prices.

Jaeden Schafer
Editor in Chief · · 5 min read
OpenAI logo

OpenAI says its coding agents can now finish 46% of the two-to-four-hour research tasks its researchers delegate to them without human intervention, up from 23% in March. The figures come from Research acceleration: The jagged frontier, a report OpenAI published November 18 on how AI agents are changing its own research work.

Longer jobs improved too. For delegated work estimated at four to eight hours of human effort, OpenAI's estimated success rate rose from 13% in March to 35% in October 2026.

The gains are uneven across task types, which is where the report gets its title. Between August 1 and October 18, agents succeeded 79% of the time when monitoring model-training runs for problems, the strongest category. Launching training runs came in at 35%, and writing inference and serving-stack code was the weakest at 25%.

Key facts

  • 01OpenAI's coding agents finish an estimated 46% of delegated 2-4 hour research tasks unaided, up from 23% in March.
  • 02Success on 4-8 hour tasks rose from 13% in March to 35% in October 2026.
  • 03Agents succeed 79% of the time monitoring training runs but only 25% writing inference and serving-stack code.
  • 04Median researcher agent use reached about $2,400 a day at API list prices in October, up 62% from September.
  • 05Documented Safety and Alignment workweeks at OpenAI rose 77% from January through September.

OpenAI cautions that rising success rates may partly reflect researchers getting better at choosing which jobs to hand off, not only stronger agents, and says the increase may reflect both. Weak categories such as serving code could also become bottlenecks on overall research gains.

Usage is climbing quickly. By October, the median OpenAI researcher used about $2,400 a day of agent activity valued at published API list prices, up 62% from September. At the 90th percentile, usage reached about $23,000 a day, up 105%. Those figures are list-price valuations, not an internal cash bill.

OpenAI attributes the growth in usage to more activity, shifts to more expensive models, and changes in agent behavior and configuration. Researchers are also running many agents at once: by mid-October, more than two-thirds of included researchers averaged a daily peak of at least four simultaneous agents, and more than 10% averaged a peak of at least 16.

To gauge how much staff value the tools, OpenAI surveyed 101 research staff between late August and early September. Asked what they would give up to restore coding agents better than GPT-5.2 if access disappeared for three months, the median answer was 30% of team compute or 40% of team headcount. The company frames these as hypothetical trade-offs, not staffing decisions.

The report also tracks where researcher attention goes. Documented Safety and Alignment workweeks at OpenAI rose 77% from January through September, which the company attributes to both hiring and researchers shifting focus. That growth picked up after OpenAI's response to the July Hugging Face incident and its work on safety cases for frontier training.

Related · from this week
Nvidia launches $4,999 DGX Spark 64GB for local AI development
Jaeden Schafer · 4 min read →

OpenAI measured that figure by classifying contributions in company-visible Slack posts, code-change requests and internal documents, then splitting each researcher's week equally across verified projects. The method estimates documented attention rather than headcount by department.

OpenAI is urging AI labs to track and publicly report indicators of recursive self-improvement. The report's call for measurement does not mean OpenAI intends to pursue that process fully.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Models

Nvidia logo
Models

Nvidia launches $4,999 DGX Spark 64GB for local AI development

The new SKU runs 100B-parameter models on device; two units cluster to 128GB and 200B via Nvidia Sync Cluster Assistant.

Jaeden Schafer4 min read
OpenAI logo
Models

OpenAI's Astra and Anthropic's Claude Opus 5 crack unsolved Enigma messages

Two cryptanalysts used frontier LLMs to break wartime codes that had resisted human researchers for two decades.

Jaeden Schafer4 min read
Anthropic logo
Models

Anthropic says Claude autonomously discovered a Crispr-like enzyme system

Nearly 950 Claude agents worked 21 hours across 210 million tokens to flag a previously uncharacterized enzyme found in bacteriophages.

Jaeden Schafer4 min read