Google’s New TimesFM-3 Outperforms Rivals At Complex Forecasting And Will Soon Be Available In BigQuery
Google has released TimesFM-3, the newest generation of its time-series foundation model and the first in the series to support native multivariate forecasting. The 330-million-parameter model is pretrained on a corpus of real-world and synthetic data exceeding one trillion time points.
It enables simultaneous prediction of multiple interrelated time series in a single forward pass without requiring task-specific fine-tuning, maintaining the zero-shot generalization that distinguished its predecessors.
Earlier TimesFM versions, including TimesFM-2.5 released in September 2025, were strictly univariate, forecasting individual series using only their own historical values. TimesFM-3 addresses the reality that most practical forecasting depends on multiple coevolving signals. The architecture supports joint prediction of several target variables, integration of historically observed covariates such as past foot traffic, and utilization of future-known dynamic inputs including promotional schedules and weather forecasts.
Under the hood, the model retains a decoder-only transformer backbone and processes data in contiguous patches of 32 timesteps with per-series normalization. Its core innovation is an alternating attention mechanism operating across a two-dimensional token grid: causal temporal attention ensures each token accesses only past data within its own series to prevent information leakage, while full variate attention enables cross-series correlation learning at every timestep.
For future-known covariates, the system concatenates current and upcoming patches into lookahead tokens. Critically, the shift to non-autoregressive inference via Contiguous Patch Masking generates the entire forecast horizon simultaneously, eliminating the latency and error accumulation associated with patch-by-patch generation in earlier versions.
Benchmark Performance and Cloud Integration
Google evaluated TimesFM-3 on three comprehensive public benchmarks—Gift-Eval, FEV-Bench, and Time—where it achieved state-of-the-art status in both point and probabilistic forecasting metrics among all pretrained foundation models. Comparisons included multivariate-capable competitors such as Chronos-2 and the Toto 2.0 family, as well as the previous TimesFM-2.5.
Even when restricted to univariate mode without cross-series or covariate information, TimesFM-3 matched or exceeded rival models; activating full multivariate operation produced substantial additional gains by exploiting inter-series dependencies. The model outputs nine quantiles spanning the 10th to 90th percentile, providing a detailed uncertainty profile rather than single-point estimates.
For enterprise deployment, the model is available on GitHub and Hugging Face. Google confirmed that BigQuery integration will arrive in the coming weeks, extending the AI.FORECAST command introduced with TimesFM-2.5 to multivariate workloads. This allows organizations in retail, finance, manufacturing, and scientific research to incorporate external signals—such as planned promotions—directly into forecasting pipelines, anticipating demand fluctuations rather than merely projecting historical patterns.
Disclaimer
In line with the Trust Project guidelines, please note that the information provided on this page is not intended to be and should not be interpreted as legal, tax, investment, financial, or any other form of advice. It is important to only invest what you can afford to lose and to seek independent financial advice if you have any doubts. For further information, we suggest referring to the terms and conditions as well as the help and support pages provided by the issuer or advertiser. MetaversePost is committed to accurate, unbiased reporting, but market conditions are subject to change without notice.
About The Author
Alisa, a dedicated journalist at the MPost, specializes in crypto, AI, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.
More articles
Alisa, a dedicated journalist at the MPost, specializes in crypto, AI, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.
