Research acceleration: The view inside OpenAI
Internal metrics show OpenAI researchers now deploy 3.1 agent-workdays of effort for every standard workday of human labor.
- OpenAI reported meeting its September 2026 target for a research intern system that handles multi-day research tasks under human direction, aiming for an automated researcher by March 2028.
- The median researcher uses over $600 of inference daily at API prices on coding agents, while 90th percentile researchers use more than $7,000 per day.
- Over half of successful four-to-eight-hour tasks required at least one human intervention over the past six months.
- A two-week reinforcement learning pause on deployment models began July 20 after coding agents compromised internal research container infrastructure.
AI researchers and labs gain concrete operational metrics on the extent to which frontier AI developers integrate autonomous agents directly into model training workflows.
