Models work with time series and tables without specific training. See applications, limits and governance for business decisions.
Direct answer
Google Research presented on October 14, 2026 TimesFM and TabFM as fundamental models for numerical distributions. They analyze time series and tabular data and are available to BigQuery users, with Gemini for business integration announced as coming. The proposal is to generate initial forecasts without specific training, such as sales or bed occupancy, but companies still need to validate horizon, error, seasonality, external variables and decision risk.
Numerical models attack another class of problem
Text alone does not represent stock, demand or capacity. Series and tables preserve quantitative relationships necessary for the operation.
Zero-shot speeds up the first benchmark
A forecast without specific training reduces the initial cost, but does not eliminate comparison with simple baseline and methods already used by the company.
Error must accompany the decision
MAPE, bias, range coverage, and period performance need to be linked to the true cost of overage, underrun, or delay.
Data perimeter does not eliminate governance
Even with data maintained in the organization's environment, access, purpose, retention, quality and auditing remain necessary.
Nexus Reading
Start with a delimited decision, create a baseline, run in parallel and only automate when the net gain is stable and explainable.
FAQ
Are TimesFM and TabFM language models?
No. They were designed for numerical distributions in time series and structured tables.
Do they require specific training?
Google says they produce initial predictions without specific training, but local validation remains mandatory.
Is the Gemini integration complete yet?
No. The October 14 announcement describes this integration as future.
Essential guides to delve deeper into the decision
This editorial analysis was produced by Nexus from the official sources below, consulted on October 19, 2026. The text is original and interprets practical implications for companies.
- Google Research — Research@ 2026: BigQuery availability, future integration, and enterprise examples.
- Google Research — TimesFM: architecture and official objective of the model for time series forecasting.
Date reported by the main source: October 14, 2026.