Thirty-two minutes, preprint unreviewed

The stamp on this IT Home page is 2026/10/23 22:43:44. It says OpenAI, PerturbAI, and Scripps Research used thousands of agents to analyze data from millions of mouse brain cells, and the analysis took 32 minutes. The team estimates that if people examined the results one by one and searched the literature, the same work could take months to a year. Thirty-two minutes is the time for this analysis. Months to a year is their estimate. It is not a timed log.

The material, in an OpenAI account dated October 21, is a CRISPR map of a whole mouse brain. The model is named GPT-Rosalind, which the page says was designed for life-science research. The map records how cell types changed after different genes were perturbed. This story records that object. How the perturbations were done is not written. The page says the work was split and run in parallel, with each agent taking one gene perturbation, the related cell response, and a literature search. The reports were gathered into a map called Reasoning Atlas, linking genes, cell changes, references, and possible mechanisms. Possible is the page's word.

The comparison numbers are in a blind review. Across 20 paired reports, biology experts preferred the distributed agent's report in 75 percent of reviews. The single agent received 16.7 percent of the preference, and 8.3 percent were ties. In the largest test, the distributed agents cited more than 700 papers, and the single agent cited 11. The 75, 16.7, and 8.3 percent figures are preferences inside those 20 pairs. Seven hundred and 11 are citation counts. The two sets do not mean the agents have already got the biology right. The page also says the distributed agents could reproduce some conclusions from existing research and offered clues that still need further checks. Reproduce means some of them. The clues are not confirmed findings.

The paper's status fixes the judgment. The preprint went on bioRxiv on October 16, 2026, and has not been peer reviewed. The team stresses that the biological links and mechanisms the agents proposed are still hypotheses waiting for experimental confirmation. The page gives specific clues. This story does not unfold those mechanisms and does not write the experimental steps. The reasoning map, the analysis flow, and the code have been opened to researchers. Opened is one sentence on the page. It is not a procedure in this story. A different item at the bottom, about nearly 2,000 agents, is not folded in.