Skip to main content

OpenAI reports higher coding-agent success and usage

19 NOVEMBER 2026·2 MIN READ·1 SOURCE

OpenAI says coding agents succeeded on an estimated 46% of its researchers’ two-to-four-hour tasks in October, up from 23% in March. Its report also puts median researcher daily agent usage at $2,400 in October.

OpenAI reports higher coding-agent success and usage

Key takeaways · 4

  • 01

    For two-to-four-hour tasks, OpenAI estimates success rose from 23% in March to 46% in October.

  • 02

    Median researcher daily usage reached $2,400 in October, 62% higher than September.

  • 03

    Success varied by task: OpenAI estimates 69% for “Communicate” and 43% for “Build.”

  • 04

    OpenAI says task delegation patterns may affect comparisons of success rates over time.

What the report measures

OpenAI says the report uses internal data to measure how coding agents have accelerated and changed its AI development cycle.[1] For tasks estimated to take two to four hours, agent success rose from 23% in March to 46% in October; for four-to-eight-hour tasks, it rose from 13% to 35%.[1] These are estimates, and OpenAI notes that researchers’ changing choices about which tasks to delegate may contribute to changes in success rates within each duration group.[1] The figures therefore describe performance on the tasks delegated, not a fixed set of work measured under unchanged conditions.[1]

Success depends on the task

OpenAI estimates a 79% success rate for tasks such as monitoring runs for issues.[1] Agents also succeeded without human intervention on 35% of training-run launches.[1] By task category, OpenAI reports estimated success of 69% for “Communicate” tasks and 43% for “Build” tasks.[1] The range suggests that a single overall success figure can obscure differences between kinds of work. OpenAI says that, despite agents completing nearly half of the measured two-to-four-hour tasks on their own, humans remain central to AI development.[1]

Usage and concurrency increased

OpenAI says median researcher daily usage reached $2,400 in October, up 62% from September; usage at the 90th percentile reached $23,000, up 105%.[1] The company attributes increased agent use to a mix of factors, including more activity, shifts to costlier models, and changes to models and agent harnesses.[1] Agent workdays per researcher rose from 0.5 to 5.7, an elevenfold increase.[1] On pretraining and posttraining teams, runtime averaged about eight workdays per researcher per calendar day.[1] By mid-October, more than two-thirds of included researchers averaged a daily peak of at least four agents working concurrently, and more than 10% averaged at least sixteen.[1]

What the report says about people

OpenAI surveyed 101 research staff from late August to early September 2026.[1] The median researcher said they would trade 40% of team headcount or 30% of current team compute, and researchers’ teams would need a median of three months to complete work equal in value to their last month’s work without coding agents.[1] Of 28 respondents, 27 chose or leaned toward OpenAI’s latest internal model.[1] OpenAI also says documented Safety and Alignment workweeks rose 77% from January to September, reflecting both new hires and existing researchers shifting toward those workstreams; growth picked up after its response to the Hugging Face incident in July 2026.[1] The company says its internal Codex, ChatGPT, and API deployment accounts for 2% of total research compute.[1] It argues that AI labs should track and publicly report indicators of recursive self-improvement, while cautioning that this reporting should not be read as an intention to pursue full recursive self-improvement.[1]

For professionals assessing coding agents, the report offers a view of both task-level performance and the usage costs OpenAI reports internally. Treat the results as a reference point, not a direct forecast: the measured task mix can change, and OpenAI says humans remain central.

Why it matters
For teams

AI training for your team

kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.

How this developed

  1. 19 November 2026

    OpenAI reports higher coding-agent success and usage

Sources

AI fluency, one session a day, built for your work.