OpenAI says its agents now do 3.1 days of work for each human research day. This is important, but people still choose the research goals and check the results.
New to this? Read it in simple words
- OpenAI says it has built an AI research intern. People still choose the research goals and check the results.
- OpenAI says its agents now do 3.1 days of work for each human research day.
- One researcher can now guide several AI tasks at once. This may change how research teams work.
- These numbers come from OpenAI itself. Outside researchers have not yet repeated the study.
- AI agent
- An AI that takes steps on its own to finish a task, such as booking or coding.
What OpenAI says it achieved
OpenAI set this goal for September 2026. The company says its AI agents now do 3.1 days of work for each human research day. The agents write code, run tests, and work on several tasks at the same time. Human researchers give them clear jobs and check what they produce.
OpenAI also says researchers are adding code faster and running more experiments than before. These numbers are useful because they describe daily work, not only a test score. However, the numbers come from OpenAI itself. Outside researchers have not yet repeated the study.
OpenAI reports 3.1 agent workdays for each human research workday. This is an early internal measure based on task length.
What the number does not prove
More agent work does not always mean more scientific progress. Research starts with choosing a good question. It also requires careful tests and a correct reading of the results. An AI agent can create more options, but a person still needs to decide which options are useful.
OpenAI says people still set research goals and decide when to continue, pause, or stop. Humans also help when an agent loses its way. This means the system is closer to a fast junior assistant than an independent scientist.
Why this still matters
One researcher can now guide several AI tasks at once. This may change how research teams work. People may spend less time writing every line of code. They may spend more time asking good questions, checking evidence, and finding errors.
OpenAI wants to build a more independent AI researcher by March 2028. Before that happens, the public needs clear measures and regular safety reports. The company has shared an early view of its progress. Independent checks are still needed.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
We used OpenAI’s report for the dates, definitions, and work numbers. We also read an outside analysis of the same data. We clearly say that the results come from OpenAI and do not prove that the AI can work as an independent scientist.