Skip to content

TimesFM

A decoder-only foundation model for time series forecasting

Visit official site
Category AI models

What TimesFM is

TimesFM is described as a decoder-only foundation model pre-trained on a large time series dataset, carrying 20 billion parameters. The material states that its zero-shot performance on multiple unseen datasets across domains and time granularities comes close to supervised methods without extra training.

What you can do with it

  • Use a model pre-trained on a large time series dataset
  • Work with a decoder-only architecture
  • Run zero-shot forecasting across domains and time granularities

Application scenarios

  • Improve demand forecasting in retail to reduce inventory costs and raise revenue
  • Carry out complex time series analysis in finance
  • Apply data forecasting in healthcare for treatment and resource planning

Who it is for

  • Teams working on retail demand forecasting
  • Analysts dealing with financial time series analysis
  • Groups handling healthcare data forecasting

Pros & cons

✓ What we like

  • Forecasts on unseen time series without extra training
  • Works across different domains and time granularities

! What to watch out for

  • The named parameter count is 20 billion, smaller than large language models
  • The described output is zero-shot forecasting on unseen datasets

FAQ

How large is the model?

The material states 20 billion parameters and a scale smaller than large language models.

Does it need training before use?

The description says it provides forecasts on unseen time series without additional training.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review