Mistral AI has released Timestral, a family of time series foundation models designed for zero-shot forecasting. The models, available in parameter sizes ranging from 5 million to 209 million, support univariate and multivariate forecasting with various covariate types and probabilistic predictions. The 209M parameter version achieves state-of-the-art accuracy on benchmarks like GIFT-Eval, TIME, and fev-bench. The source code is available on GitHub under a modified MIT license. AI
IMPACT Provides a new foundation model for time series forecasting, potentially improving accuracy and efficiency in applications requiring predictive analytics.
RANK_REASON Model release from a frontier lab (Mistral AI) with system card details. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Hugging Face Trending Models →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
