Google AI model ranks first among 39 CDC flu forecasts
Google’s ERA-built model ranked first among 39 eligible submissions in the CDC’s evaluation of US flu-hospitalisation forecasts for the 2025–26 season. The result offers a public test of AI-generated forecasting software, but the CDC’s combined forecast—used to inform expectations of hospital demand—ranked seventh overall.
Bottom line — The CDC says 33 of 39 models beat its baseline, while Google’s ERA remains available only to trusted testers.
Go deeper 5
-
The CDC report says 34 teams submitted 53 models; 39 met the evaluation’s submission threshold.
-
Google says ERA uses AI to generate optimisation algorithms; a research paper describes a language model guiding a search that writes, tests and refines forecasting software.
-
The CDC’s FluSight ensemble combines eligible model forecasts and ranked seventh on the season’s primary metric; Google’s was the top individual team submission.
-
The CDC characterised the season as moderate, with weekly hospital admissions exceeding 40,000 at the peak in the week ending 27 December 2025.
-
The CDC says the combined forecast’s prediction intervals missed the late-December rise and mid-January fall; fewer than 25% of two-week-ahead intervals across jurisdictions captured observed values around the peak.