Google TimesFM-3 tops forecast benchmarks, non-commercial
TL;DR
- TimesFM-3 is the first Google time-series model natively pre-trained for multivariate forecasting rather than retrofitted post-hoc onto a univariate architecture.
- Model weights ship under timesfm-non-commercial-license-v1.0; the source code is Apache-2.0, but production deployment of the default pretrained weights is explicitly prohibited.
- TimesFM-2.5 remains the only Google time-series checkpoint cleared for commercial or production use, a distinction most launch coverage buries.
Google Research on Monday released TimesFM-3, a 330-million-parameter decoder-only foundation model that forecasts multiple time series jointly and takes the top average rank on three major zero-shot benchmarks. It ships under a non-commercial license that keeps it out of production systems.
The pitch is that forecasting in the wild is rarely one series in isolation. "Most real-world forecasting problems are inherently multivariate: where multiple time series and auxiliary external features jointly impact the future forecast of a time series," authors Ayush Jain and Rajat Sen write. Earlier versions could not do it. "Up until TimesFM-2.5 (released in September 2025), our models were strictly limited to univariate forecasting." The new release adds native support for multiple targets, past covariates, and past-future dynamic covariates such as promotion schedules or weather forecasts.
The architectural shift swaps the iterative decoder for what the team calls "Contiguous Patch Masking to generate the entire forecasting horizon in a single forward pass." Patches of 32 time steps run through alternating causal temporal attention (within-series) and full variate attention (cross-series), producing 9 quantiles per step for probabilistic forecasts.
The headline claim, per the blog: "On all three benchmarks, TimesFM-3 is the top-ranked model in terms of both point and probabilistic forecasting metrics among all pre-trained foundation models." MarkTechPost breaks that down as rank #1 on fev-bench across 100 real-world tasks and rank #1 on TIME across 50 domain datasets and 98 evaluation tasks, ahead of Chronos-2 and the Toto 2.0 family.
The catch is the license. As The New Stack headlined it, "You can't use it at work (yet)." The Hugging Face weights "are restricted to non-commercial, non-production use," while the code repository stays Apache-2.0. That puts TimesFM-3 in a different bucket from the permissively-licensed drops we've been tracking on the open source beat (255 alerts in the last 90 days): fair game for benchmarking today, not for shipping behind a forecast API. BigQuery integration is promised "in the coming weeks."
What others are reporting
-
The New Stack Read →
Frames the story as a deployment-access gap: benchmark superiority is real, but the non-commercial license blocks enterprise adoption and the headline makes that the lede.
Google's new forecasting model beats everyone. You can't use it at work (yet).
-
TSFM.ai Read →
Provides architectural depth and benchmark skepticism: deconstructs how average rank obscures effect size and flags potential pretraining overlap, plus practical evaluation guidance for practitioners.
TimesFM 3 is pretrained to distinguish three semantic roles: Targets, Past-only covariates, and Past-future covariates.
-
MarkTechPost Read →
Names all three evaluation frameworks and clarifies the dual-license split (Apache-2.0 code vs. non-commercial weights), making the commercial-deployment blocker explicit and actionable for readers evaluating deployment.
TimesFM-3 is pretrained natively for multivariate forecasting on more than 1 trillion time points, and accepts multiple targets.
-
GitHub (google-research/timesfm) Read →
First-party source confirming non-commercial weight restriction verbatim, 16k context support, and rank-1 across fev-bench, TIME Benchmark, and GIFT-Eval — the authoritative technical ground truth.
Model weights (3.0): restricted to non-commercial, non-production use under timesfm-non-commercial-license-v1.0.
Originally reported by research.google
Read the original article →Original headline: Google Research Ships TimesFM-3, a 330M-Param Zero-Shot Foundation Model for Multivariate Forecasting