TimesFM-3

TimesFM-3 is an AI model built by Google Research that looks at several kinds of data together to forecast future numbers. Say an ice cream shop wants to predict tomorrow's sales — the old approach would just look at the past sales chart and extend the day-of-week pattern forward. But in reality, there's often information that's already known in advance and will affect sales, like a discount event happening next week. TimesFM-3 feeds that kind of scheduled information into the model alongside the sales history, calculates how the two interact, and produces a more accurate forecast.

To put it another way, if an older model is like someone who decides what to wear based only on the weather forecast, TimesFM-3 is more like someone who checks the weather forecast and also looks at the day's outdoor event schedule before picking an outfit. Because it weighs multiple pieces of information at once, it can make finer-grained predictions, such as how much extra a discount event will boost sales on a given day.

It also doesn't just spit out a single number as its forecast. It shows a range from the lowest to the highest likely value, so you know how much to trust the prediction. It was built with practical use cases in mind — fields like retail, finance, and healthcare, where looking ahead into the future matters.

In articles, it shows up in sentences like "Google Research released TimesFM-3, a foundation model for multivariate time-series forecasting." Here, "multivariate" means the model doesn't just look at sales alone — it factors in multiple data streams at once, such as promotion schedules or sales of related products. The "3" after the name simply marks it as a newer version than its predecessor, TimesFM-2.5 — it doesn't refer to a separate, different product.