Google Research releases TimesFM-3 multivariate forecasting model

Google Research has released TimesFM-3, the third generation of its time-series foundation model family. Unlike its predecessors, which were limited to univariate forecasting, TimesFM-3 is natively pretrained for multivariate forecasting. This allows the model to predict multiple related signals in a single forward pass, enabling different data points to inform one another.

The 330-million-parameter model was pretrained on a corpus of more than 1 trillion time points, consisting of both real-world and synthetic data. It is designed for zero-shot generalization, meaning it can handle complex multivariate scenarios—such as forecasting multiple related products or incorporating past covariates—without requiring task-specific fine-tuning. This capability aims to replace hand-engineered pipelines in sectors including retail, finance, manufacturing, healthcare, and natural sciences.