Skip to content

Chronos

Pretrained language models for time series forecasting

Visit official site
Category AI models

What Chronos is

The model is described as a family of pretrained time series forecasting models based on a language model architecture, where values are scaled and quantised into tokens.

What you can do with it

  • Turn a time series into a sequence of tokens
  • Train a language model with cross-entropy loss
  • Sample multiple future trajectories for a forecast

Application scenarios

  • An energy company forecasts power demand
  • A financial analyst forecasts stock price movements
  • A retailer forecasts sales to tune inventory

Model notes

  • Forecasts are described as probabilistic
  • Training uses large public time series data
  • Synthetic data is generated with Gaussian processes

Pros & cons

✓ What we like

  • Multiple future trajectories are sampled from history
  • Training covers public and Gaussian-process synthetic data

! What to watch out for

  • Forecasts are probabilistic rather than single values
  • The approach depends on scaling and quantisation

FAQ

How are forecasts produced?

The material describes sampling multiple future trajectories from a historical context.

What data was used for training?

Large public time series data and synthetic data from Gaussian processes are named.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review