TimesFM

lastmod 2026-04-02created 2026-04-022 edits

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

Google's 200M-parameter, 16k-context time-series foundation model

About

• TimesFM is a large pretrained model for time-series forecasting developed by Google Research, using a decoder-only architecture
• The latest version, 2.5, supports 200 million parameters, a 16k context length, and forecasting up to a 1k horizon
• Includes a variety of features such as continuous quantile forecasting, covariate input (XReg), and Flax and PyTorch backends
• The model is released through Hugging Face and is also integrated into BigQuery
• A general-purpose foundation model for time-series forecasting that offers scalability and efficiency for both research and practical use

See also

• Time series
• Deep learning
• Google

Favorite site

• TimesFM - Google's 200M-parameter, 16k-context time-series foundation model | GeekNews
• Github - google-research/timesfm
• A decoder-only foundation model for time-series forecasting