Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

OpenAI agents double task-completion rate since March, hit 46% on mid-length research work

46% autonomous success on 2-to-4-hour tasks, up from 23% in March.

OpenAI's internal coding agents now complete 46% of research tasks estimated to require two to four hours of human work, up from 23% in March, the company reported on November 191. For longer tasks estimated at four to eight hours, the autonomous success rate rose from 13% to 35% over the same period.

The update, which OpenAI framed as a progress report on research acceleration, also disclosed that the median OpenAI researcher's daily agent usage reached $2,400 at API list prices in October, a 62% increase from September. At the 90th percentile, daily usage hit $23,000, up 105% from September.

OpenAI described the productivity gains as uneven across task types. The autonomous agent success rate was 79% for monitoring training runs, 35% for launching training runs, and 25% for writing code for inference and serving stacks. The company characterized the pattern as a "jagged frontier," noting that humans remain central to AI development and "a core part of the AI research loop".

The gap between median and heavy users is widening. The 90th-percentile researcher's $23,000 daily spend is nearly ten times the median's $2,400, and the month-over-month growth rate at the top (105%) outpaced the median (62%).

ANALYSIS The spending distribution suggests a small cohort of researchers is integrating agents far more deeply into their workflows than the typical user, a pattern consistent with OpenAI's own observation that "the heaviest users are pulling further ahead."

The disclosure arrives the same day Anthropic began trading on Nasdaq at $111.50 per share, 34% above its IPO price[2], and one day after Google DeepMind released Gemini 4 Neon[3].

ANALYSIS By publishing concrete internal metrics on agent-driven research productivity, OpenAI is staking a public claim on the recursive-improvement narrative at a moment when both its closest competitors are making high-profile moves of their own.

OpenAI's report did not disclose the total number of researchers included in the measurements or the absolute volume of tasks evaluated.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.