OpenAI Coding Agents Post 46% Success Rate on Two-to-Four-Hour Tasks
The rate for four-to-eight-hour tasks rose to 35% from 13%, while researchers' median daily agent use was valued at $2,400 in October.

OpenAI coding agents recorded a 46% autonomous success rate in October on research tasks estimated to require two to four hours, compared with 23% in March.
For tasks estimated at four to eight hours, the success rate rose to 35% from 13% over the same period. The October rate for that longer task band was 11 percentage points below the rate for two-to-four-hour tasks.
The median daily agent use among researchers was valued at $2,400 in October at API list prices, up 62% from September. Daily use at the 90th percentile was valued at $23,000, an increase of 105%.
Recorded workweeks in safety and alignment increased 77% from January through September, the largest absolute increase among the major research fields tracked. OpenAI called on AI labs to track measures of recursive self-improvement (RSI) and publicly disclose them.