Skip to content
OpenAI·Indexed by this site · Originally published· Evaluation →· Its place in the scan record →SelectedAttention74

OpenAI releases its second research acceleration report: agent workdays per person per day rose from 0.5 to 5.7

Research acceleration: The jagged frontier

AI summary

OpenAI releases its second research acceleration report, using internal data to quantify how coding agents accelerate its AI research. At API list prices, the median researcher agent usage reached $2,400 per day and the 90th percentile about $23,000, up 62% and 105% from September; agent workdays per person per day rose from 0.5 to 5.7. From August 1 to October 18, 2026, the estimated success rate without human intervention was 69% for Communicate-type tasks and 43% for Build-type tasks, and only 25% for writing inference and serving stack code. Among 101 surveyed research staff, the median was willing to trade 40% of team headcount or 30% of team compute to get frontier coding agents back; researchers' time investment in safety and alignment has nearly doubled since the start of the year, the largest absolute increase of any research direction. OpenAI says AI labs should be required to track and publicly report recursive self-improvement (RSI) metrics, and stresses that this report does not represent an intention to pursue full RSI.

Why it was selected

OpenAI publishes internal measurements including agent usage, task success rates and a capability index, providing a quantitative basis for tracking research acceleration and discussing recursive self-improvement governance.

Read originalopenai.com