OpenAI says it has reached a milestone it set for an automated research intern: tasks that, under human direction, are well-defined and would take a skilled researcher a few days. The company says the goal, set in fall 2025, was reached in September 2026.

The more revealing number is workload rather than a launch label. According to OpenAI’s measurements, its systems handled about 3.1 agent-workdays for every human workday by mid-August. OpenAI says the figure is an internal measurement, not an industry benchmark. At API prices, it says the median researcher consumed more than $600 a day in inference, while the 90th percentile exceeded $7,000 a day.

Those figures do not mean the systems are working unsupervised. OpenAI says more than half of successful tasks lasting four to eight hours had at least one human intervention in the previous six months. The company’s chart also warns that these metrics are hard to interpret, that progress may not keep pace, and that the results are preliminary. The measurements describe a fast-moving workflow, not a verified claim of general autonomy.

That qualification is central to the story. The same-day picture OpenAI presents is of agents taking on increasingly expensive, multi-hour research work while people still set the task, intervene and judge the result. Transparency about inference cost, intervention rates and task boundaries is therefore as important as the headline productivity ratio.

In a separate September 6 post, OpenAI chief scientist Jakub Pachocki wrote that he does not believe any lab has solved alignment and monitoring well enough to keep scaling responsibly at maximum speed for much longer. He said he expects and hopes for voluntary slowdowns until shared safety bars exist. That is a request for coordination and disclosure, not evidence that a technical alignment solution is finished.

OpenAIs research post also places a fully automated AI researcher in March 2028 as a target or belief. That date is not verified here and should not be read as a forecast. The safer conclusion from the company’s own account is narrower: agent-workday output is rising, the costs can be substantial, humans remain in the loop, and OpenAI is pairing the productivity claim with a public warning about safety limits.

Sources: OpenAI, Research acceleration: A view inside OpenAI” (https://openai.com/index/research-acceleration-view-inside-openai/); Jakub Pachocki, “An Alien Mind” (https://openai.com/index/an-alien-mind/).