Google TimesFM 3.0 Targets Practical Multivariate Forecasting

Google released TimesFM 3.0, a 330M-parameter zero-shot forecasting model with native multivariate prediction, roughly trillion-point pretraining, uncertainty outputs, and support for known-future signals such as promotions, holidays, and weather. Reported benchmark gains are promising for planning and demand forecasting, but non-commercial weights and the lack of independent validation limit immediate production significance.