Back to Model
T

Timestral

Model Oct 15, 2026

Mistral's downloadable time-series forecasting family, with CPU or GPU inference and 5M, 28M, 123M and 209M parameter checkpoints.

Visit primary link 3 sources attached
Mistral's six-panel chart comparing the Timestral family's parameter counts and reported forecasting metrics with other public models.
Mistral's model-size and forecasting comparison from its English model card. These are author-reported benchmark results, not independently reproduced by AITrending. Source: Mistral AI
self-hosted

What happened lately

Timestral timeline

Mistral announced Timestral on October 15, 2026, as its first family of time-series foundation models. Four checkpoints cover 5M to 209M parameters. The model predicts numerical series and uncertainty quantiles, with support for multiple target series, past-only covariates and known-future covariates. It does not itself provide a conversational agent or operate infrastructure.

The standalone inference package requires Python 3.10 or later and PyTorch 2.4 or later, and can run on a CPU or GPU. Its Apache-2.0 code license does not cover the downloadable weights. Their Modified MIT license excludes companies or employers exceeding $20 million in global consolidated revenue during the preceding month; those organizations must seek a separate commercial license or use Mistral’s hosted services. Performance claims remain Mistral’s measurements.