OpenAI's Coding Agents Now Finish 46% of Medium Research Tasks
OpenAI's internal report says agents now finish 46% of 2-4 hour research tasks with zero intervention, while median daily agent spend hit $2,400 and parallel agent fleets became standard.
By Grace Kim
4 min read

What's News
- Zero-intervention success on 2-4 hour research tasks rose from 23% in March to 46% in October 2026, per OpenAI's November 18 report.
- Median researcher agent spend hit $2,400 per day in October, up 62% month-over-month; the 90th percentile hit $23,000.
- Agent workdays per researcher rose elevenfold, from 0.5 to 5.7 per calendar day.
- Autonomous success ranged from 79% for monitoring training runs to 25% for writing inference and serving code.
- 101 surveyed researchers said they'd trade a median 40% of team headcount to keep frontier coding agents for three months.
OpenAI's coding agents autonomously completed 46% of research tasks that would take humans two to four hours, up from 23% in March, according to an internal report published November 18. The report, titled "Research acceleration: The jagged frontier," draws on company tracking from March to October 2026 and marks OpenAI's most concrete public benchmark yet for AI-driven research acceleration.
The doubling comes with caveats. For tasks estimated at four to eight human hours, zero-intervention success climbed from 13% to 35%. OpenAI cautions that these duration buckets mix capability gains with researchers' shifting choices about what to delegate — as agents improve, staff hand them different tasks within the same band, so the rates don't isolate model improvement alone.
How much do results vary by task type?
Sharply. Using the O*NET AI R&D taxonomy developed by Epoch AI, OpenAI measured autonomous success rates from August 1 to October 18 across specific activities:
- Monitoring training runs: 79%
- Answering technical questions: 72%
- Launching training runs: 35%
- Building evaluation and RL environments: 29%
- Writing inference and serving code: 25%
OpenAI notes the lower-performing categories could become bottlenecks to overall uplift.
What is this costing?
Usage and spend are climbing fast. At API list prices, the median researcher's daily agent spend hit $2,400 in October, up 62% from September. The 90th percentile reached $23,000, up 105%. Measured in "agent workdays" — eight hours of agent runtime — the average rose elevenfold, from 0.5 to 5.7 per researcher per calendar day.
Parallelism is now standard. By mid-October, more than two-thirds of researchers ran at least four agents concurrently at their daily peak; over 10% ran sixteen or more. Internal agent spend in September ran roughly twice the non-equity compensation bill for researchers, though equity makes up 70–80% of target pay. Agent compute still accounts for only 2% of total research compute.
What do researchers actually think?
A survey of 101 research staff, conducted from late August to early September, asked what they'd trade to keep frontier coding agents for three months. The median answers: 40% of team headcount or 30% of team compute. Researchers also estimated they'd need three months to match one month of agent-assisted output using only GitHub Copilot plus ChatGPT. OpenAI flags the survey's ad-hoc sampling and the unreliability of self-reports.
Follow-up interviews with ten researchers described agents as strong at implementation, debugging and large-scale scanning, but weak on research taste, code quality, data quality and writing. Agents "rabbit-holed on unnecessary steps," produced "AI slop" pull requests, and made invalid analysis assumptions. One post-training researcher described the required workflow: "Go make one. I'll review it and make edits. Then go do the rest."
A pretraining researcher put it more bluntly: "I do think I have more taste in terms of the ideas. But I have no confidence that will continue to be true for any significant period of time."
Where do the ideas still come from?
Humans, mostly. OpenAI asked key contributors to GPT-6 Astra and GPT-6.2 Astra where their most important ideas originated. For both releases, post-training and safety researchers said people generated all the ideas, with models helping implement them. Pretraining researchers credited people for GPT-6 Astra. For GPT-6.2 Astra, two pretraining researchers described developing ideas in conversation with Codex — one said Codex identified key insights, another said a Codex suggestion made the main idea work.
What's rising fastest?
Safety work. Documented researcher workweeks in Safety & Alignment almost doubled from January to September, the largest absolute increase of any research area. OpenAI attributes this to new hires and staff shifting focus, with an inflection after the "Hugging Face incident" in July 2026.
Output metrics are up too. Lines of code changed per active contributor and experiments per active experimenter — 1.68× the 2025 baseline in the four weeks to October 18 — have risen sharply. OpenAI also introduced an internal version of Epoch AI's Capabilities Index (OECI): its frontier for OpenAI models has risen 16.4 points per year since August 2025, versus the 14 points per year Epoch observed since late 2024. OpenAI notes its models score higher on OECI than on Epoch's public ECI, likely due to benchmark mixture differences.
OpenAI stresses that historical human-AI collaboration trends may not predict fully autonomous progress, and that its methods — classifiers on Slack posts, pull requests and internal documents — proxy attention allocation, not formal assignments. The company says publishing these metrics is not a statement of intent to pursue full recursive self-improvement (RSI), but a call for industry-wide RSI reporting norms, citing three Anthropic reports, a Google DeepMind section in the Gemini 4 Argon safety report, and a METR-Anthropic collaboration.
The 46% figure is a specific, testable claim about one task band inside one lab. Whether that velocity translates into sustained capability acceleration — and whether the jagged frontier smooths or deepens — is what the OECI and future reports will test.
Source: epyhia.com
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,582 articles