OpenAI, PerturbAI and Scripps Research say they used close to 2,000 AI agents to analyze a large mouse brain dataset in 32 minutes.
OpenAI published the work on Wednesday. The agents ran on GPT-Rosalind, a model that appears aimed at the life sciences. The team estimates that researchers working by hand would need months to a year for the same analysis. Both numbers come from the team itself: the 32 minutes is its reported figure, and the by-hand timeline is an estimate, not a head-to-head test.
Take the claim at face value and the gap is large. A single month is about 43,000 minutes, so even the low end of "months" implies a speedup of well over a thousand times.
The source does not say how the work was divided. The agent count and the clock suggest the agents ran in parallel, each taking a slice of the dataset, but that is our inference, not something the team has described. If so, the speed comes mostly from splitting up the work, not from any single agent being fast.
The harder question is quality. Nothing in the material we reviewed explains how the output was checked against what human analysts would have found. In biology, a fast answer only helps if researchers can trust it, and the checking may take longer than the computing. That is the number to look for when more detail appears.