saved
TimesFM-3: A zero-shot foundation model for multivariate forecasting
gist
Google Research scientists Ayush Jain and Rajat Sen introduce TimesFM-3, a 330-million-parameter time-series foundation model pretrained natively for multivariate forecasting on more than a trillion time points. Unlike TimesFM-2.5, it jointly predicts coevolving series plus past and past-future covariates in one forward pass, using alternating causal temporal and full variate attention and Contiguous Patch Masking. It ranks first among pretrained foundation models on Gift-Eval, FEV-Bench, and Time for both point and probabilistic metrics, even in univariate mode. Weights are on GitHub and Hugging Face; BigQuery integration is coming.
ideas
- Univariate history is not enough. Through 2.5, TimesFM forecast from a single series' history; TimesFM-3 is pretrained natively to jointly forecast related targets with past and known-future covariates.
- Attention is a 2D grid. Causal temporal attention stays within one series; full variate attention at a timestep sees every other series so promotions in one can move sales in another.
- The whole horizon lands in one pass. Contiguous Patch Masking fills masked future patches at once, avoiding autoregressive error and latency, and emits nine quantiles per target step.
- Even univariate mode already leads. On Gift-Eval, FEV-Bench, and Time it is top-ranked among pretrained foundation models for point and probabilistic metrics; multivariate mode is a further leap.
- Weights now, BigQuery later. TimesFM-3 is on GitHub and Hugging Face; BigQuery AI.FORECAST still points at TimesFM-2.5 until the new integration lands.
quotes
“Up until TimesFM-2.5 (released in September 2025), our models were strictly limited to univariate forecasting”
“TimesFM-3 uses the strategy of Contiguous Patch Masking to generate the entire forecasting horizon in a single forward pass.”
“TimesFM-3 is the top-ranked model in terms of both point and probabilistic forecasting metrics among all pre-trained foundation models.”
“These two attention mechanisms alternate for several layers, seamlessly blending temporal patterns with cross-series relationships.”