Google Research released TimesFM 3.0 in August 2026, a pretrained time series foundation model that can forecast data it has never seen with no extra training. The model has 330 million parameters and was pretrained on a mix of real and synthetic data containing more than 1 trillion time points, allowing it to forecast new series "zero-shot." Its most important upgrade is native multivariate forecasting: earlier versions, up to TimesFM 2.5 released in September 2025, could only look at one series at a time, while TimesFM 3.0 can read several related series together and use outside signals like promotions, weather, and holidays. TimesFM stands for Time Series Foundation Model, and the model is a decoder-only transformer for time series. It is available on BigQuery, GitHub, and Hugging Face. The article notes that traditional tools such as ARIMA or exponential smoothing needed a specialist to pick settings for each dataset, and machine learning models often needed lots of historical data, feature engineering, and retraining whenever conditions changed—slow and costly for companies with thousands of products, stores, or sensors.