OpenAI: automated research intern milestone hit — 3.1 agent-days per human day

On 6 Sep 2026 OpenAI published “Research acceleration: The view inside OpenAI,” claiming it has met the goal announced last fall of an automated research intern by September — a system that can carry out well-defined research tasks under human direction, including multi-day work for a skilled researcher — and that it is aiming for a more capable automated AI researcher by March 2028. Internal metrics: by mid-August the median researcher used more than $600/day of agent inference at API prices (90th percentile >$7,000/day), and the research org now runs 3.1 agent-workdays per human workday (up from below 1× before June). Agents still often need human intervention on longer tasks. The post ties recent Hugging Face and Astra Critical-cyber security pauses into the acceleration story, noting RL pauses and hardened sandboxes while researchers shifted compute across model classes. Simon Willison (6 Sep) highlighted the same OpenAI research-acceleration post for a broader developer audience.