Chronos-2: Zeroshot Multivariate Forecasting with Covariates — A Practical Tutorial for Retail Forecasting

Chronos was among the first models to ignite the foundation-model wave in time-series forecasting.

Last year brought its second upgrade, Chronos-Bolt, with notable performance gains. This month, the team released Chronos-2—the 3rd and most capable variant yet.

Chronos-2 not only outperforms its predecessors but also adds longer context windows and external covariate support.

We’ll break down Chronos-2’s architecture in a follow-up. Here, we’ll test it against strong statistical baselines and compare it to TimesFM-2.5 under the same setup.

Let’s get started!

✅ Find the full notebook for this article here: AI Projects Folder (Project 26)

Chronos-2 supports 4 forecasting modes:

• Univariate

• Multivariate

• Univariate + Covariates

• Multivariate + Covariates

Multivariate forecasting jointly predicts all time series, capturing their interdependencies. It works best when the series are complex and share similar characteristics.

Chronos-2 also supports finetuning.