Mistral releases Timestral, new time-series forecasting models

Mistral AI has released Timestral, a family of foundation models designed specifically for time-series forecasting. The release includes four model sizes ranging from 5 million to 209 million parameters, designed to be efficient enough to run on standard CPUs while delivering state-of-the-art probabilistic forecasting accuracy.

The models are built for zero-shot use, meaning they can make predictions without task-specific fine-tuning. Time-series forecasting—predicting future values based on historical patterns—is critical for applications in finance, energy, supply chain, and other domains where trends matter.