OpenAI reports that its coding agents now complete nearly half of two-to-four-hour research tasks unaided
In a report dated November 18, OpenAI shares internal measurements of how coding agents are changing its AI research, as an update to its September report. It calls the uplift "jagged" and says humans remain central to the work.
- Task success: OpenAI estimates that agents' success without human intervention rose from March to October from 23% to 46% on tasks that would take a human two to four hours, and from 13% to 35% on four-to-eight-hour tasks. By task type it puts success at 79% for monitoring training runs, 72% for answering technical questions, 35% for launching training runs, 29% for building evaluation and agent environments and 25% for writing inference and serving code.
- Usage: at API list prices, the median researcher's daily agent use reached $2,400 in October (up 62% from September) and the 90th percentile $23,000 (up 105%). Agent workdays per researcher rose from 0.5 to 5.7. OpenAI says the API-equivalent value of agent use was about twice researchers' non-equity pay as of September, while internal deployment takes 2% of research compute.
- Survey: in a survey of 101 research staff, the median would give up 40% of team headcount or 30% of team compute to keep frontier coding agents for three months. In a smaller survey, 27 of 28 respondents expected to be more productive with the latest internal model than with a mid-career research engineer plus a 2024 model.
- Ideas: researchers said people originated the key ideas behind GPT-6 Astra and GPT-6.2 Astra, with models helping implement them. Some pretraining researchers said Codex helped develop ideas for GPT-6.2 Astra.
- Safety: OpenAI says documented Safety and Alignment work rose 77% from January to September, the largest increase of any research area, picking up after the Hugging Face incident in July.
OpenAI says the report is meant to support public tracking of indicators of recursive self-improvement, and that it should not be read as an intention to pursue it. The figures are OpenAI's own estimates, and it notes that its task-delegation and self-report data have limits.
Sources
Get news like this as a push.
AI Tracker alerts you when something changes for the tools you use.