Timestral: Mistral release time-series foundation models

Mistral has released Timestral, its first family of time-series foundation models, with the largest version containing just 209M parameters and ranking first on all three major forecasting benchmarks (GIFT-Eval, TIME, and fev-bench) while outperforming larger competitors. The models, available in four sizes from 5M to 209M parameters, support probabilistic multivariate forecasting with covariates and can run efficiently on standard CPUs without requiring GPUs. Timestral uses synthetic pre-training and covariate-aware attention to enable zero-shot forecasting across diverse domains including industrial, financial, and energy sectors.