Mistral releases Timestral time-series foundation models up to 209M parameters

Timestral provides multivariate, probabilistic forecasting using covariate-aware attention and synthetic pre-training. The 209M parameter model outperforms larger existing models on GIFT-Eval, TIME, and fev-bench benchmarks. The model family scales from 5M to 209M parameters and supports zero-shot forecasting and fine-tuning via Mistral Forge.