<ul data-eligibleForWebStory="false"><li>BigQuery now integrates TimesFM, a pre-trained forecasting model from Google Research, to enable accurate time-series predictions for various business applications.</li><li>TimesFM is a decoder-only transformer model trained on 400 billion real-world time points, providing 'zero-shot' forecasting capabilities without the need for additional training.</li><li>TimesFM 2.0, with 500 million parameters, is now native to BigQuery, offering fast and scalable forecasting of millions of time series with a single SQL query.</li><li>For quick and general forecasting needs, TimesFM is recommended, while ARIMA_PLUS is more suitable for fine-tuning, explainable results, and leveraging longer historical data.</li></ul>

Zero-shot forecasting in BigQuery with the TimesFM foundation model

Discover more