TimesFM-3

TimesFM-3, released August 2026 by Google Research, is a 330-million-parameter time-series foundation model pretrained on over 1 trillion real and synthetic time points. Unlike every TimesFM checkpoint through 2.5, which were univariate, TimesFM-3 is natively multivariate: it forecasts multiple related series in a single forward pass, accepting multiple targets, past covariates, and past-future covariates with no task-specific fine-tuning. Google reports top average rank among pretrained foundation models on GIFT-Eval, fev-bench, and the TIME leaderboard.

The honest picture: the TimesFM-3 weights ship under a non-commercial license — you can benchmark but not run them behind a production forecasting API — a sharp downgrade from the Apache-2.0 lineage of earlier versions. At 330M it is also meaningfully heavier than 2.5's 200M, and multivariate zero-shot quality on messy real-world series still needs independent validation.