Model Comparison

Chronos vs Moirai: architecture, context, and trade-offs

Amazon Chronos tokenizes time series into a T5 encoder-decoder. Salesforce Moirai uses a universal transformer that handles multivariate series and covariates natively. The right pick depends on your data shape, context budget, and whether you need exogenous variables.

Head-to-head benchmarksCovariate support comparisonTry both on your data

Compare in playground View all models

Side by sideSame request, different models

# Chronos
{"model": "amazon/chronos-bolt-base", "inputs": [...], "parameters": {"prediction_length": 14}}

# Moirai
{"model": "salesforce/moirai-1.1-r-base", "inputs": [...], "parameters": {"prediction_length": 14}}

Swap the model ID. Everything else stays the same.

Chronos

T5-based encoder-decoder, mini through large, strong probabilistic forecasts

Moirai

Universal transformer, native multivariate + covariate support, up to 5000 context

On TSFM.ai

Both available through one API with identical request shapes

Feature comparison

Dimension	Chronos	Moirai
Architecture	Encoder-decoder transformer (T5-based)	Universal forecasting transformer
Model sizes	Mini, Small, Base, Large	Small, Base, Large
Parameters	8M to 710M across sizes	14M to 311M across sizes
Context length	Up to 2048 time steps	Up to 5000 time steps
Covariate support	Univariate only — no exogenous variables	Native multivariate with covariates (past + future)
Quantile support	Native quantile generation via tokenized distribution	Flexible distribution head with mixture outputs
Pre-training data	Synthetic + real-world time series (TSMixup augmentation)	LOTSA: 27B observations across 9 domains
Typical latency	50–200ms depending on size	60–250ms depending on size and context

When to choose which

Choose Chronos when

Your workload is univariate, you want fine-grained control over latency vs accuracy trade-offs through model size selection, or you need the proven reliability of the T5 architecture with native probabilistic outputs.

Choose Moirai when

You need to include covariates or exogenous variables, your series are multivariate, you want longer context windows up to 5000 steps, or your data spans multiple domains and frequencies that benefit from Moirai's any-variable, any-frequency design.

Or compare both

Send the same series to both models through the TSFM.ai API. Compare forecasts, quantiles, and latency on your own data before committing to either.

Try in playground

Compare on your own data

The best model comparison is the one you run on your own series. Try both Chronos and Moirai in the playground.

Open playground Get API key

Chronos vs Moirai: architecture, context, and trade-offs

Feature comparison

When to choose which

Choose Chronos when

Choose Moirai when

Or compare both

Frequently Asked Questions

Related Pages

Chronos vs TimesFM

TimesFM vs Moirai

Best time series foundation models in 2026

Compare on your own data