OpenAI released internal data showing AI agents now perform three times the work of human researchers, with a 124-fold increase in token output since December 2025. It is the company's first major update on AI research automation since the launch of GPT-6 Astra in September 2026.
The company reported a median researcher at OpenAI spends over $600 a day on inference at API prices, with the 90th percentile exceeding $7,000. That compares with lower costs in other parts of the company, where agent runtime has topped human working hours since June.
The AI agents are built on internal research frameworks and target automating research tasks under human guidance. Availability of the automated systems began in mid-August, initially for internal researchers and development teams.
"The most important control tool is losing its edge," said Jakub Pachocki, chief scientist. "I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence." Pachocki warned that current control tools are becoming less reliable as models grow more complex.
The announcement follows the unveiling of GPT-6 Astra, which the company says is better aligned than its predecessor. OpenAI did not say how it will measure progress on alignment and monitoring, and raised concerns about whether control mechanisms can keep pace with AI growth.
Source: thedecoder
