Skip to content
Hot eventDeveloping

OpenAI reports uneven gains from coding agents in AI research

2 reports2 reporting sourcesUpdated 6 h ago

Start here

AI digest

OpenAI has published internal data on how deeply coding agents are embedded in its own model development, and says their performance across research tasks remains uneven — a pattern it calls a "jagged frontier." The company frames the results as relevant to recursive self-improvement, in which AI systems speed up AI development itself. According to the figures, the estimated success rate of autonomous agents on two- to four-hour tasks rose from 23% to 46% between March and October, but agents still failed often on complex work: 29% success at building evaluation and reinforcement-learning environments and 25% at writing inference and serving code. At regular API list prices, the median researcher would have consumed $2,400 worth of tokens per day in October, up 62% from the previous month, while internal agent use accounted for just 2% of the company's total research compute. In an internal, non-representative survey of 101 employees, researchers said at the median they would rather give up 40% of their team's staff or 30% of their compute capacity than go three months without current coding agents. OpenAI cautions that data from human-AI collaboration so far may not extrapolate linearly.

Generated from reports · updated 6 h ago

Report timeline

Follow the reports to see the event from different sides.

November 19
  1. The Decoder
    OpenAI researchers would rather lose 40% of teammates than give up coding agents, internal report says

    7 workdays per researcher per calendar day. The estimated success rate of autonomous agents on two- to four-hour tasks rose from 23 to 46 percent between March and October, but agents still fail often at complex work, succeeding only 29 percent of the time at building evaluation and reinforcement-learning environments and 25 percent at writing inference and serving code.

November 18
  1. OpenAI News
    OpenAI Research Acceleration: The Jagged Frontier of Coding Agents in AI Research

    OpenAI has measured how coding agents accelerate AI research, finding their performance remains uneven across tasks. The work examines what this jagged frontier signals about recursive self-improvement.

Heat for this event

Current, comparable range
12
Peak, comparable range
14November 19 22:00
Change in 24 hours
–

The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.