Mistral AI has released Timestral, a new family of time series foundation models. Available in parameter versions ranging from 5 million to 209 million, Timestral is designed for zero-shot forecasting. The models support both univariate and multivariate time series, can handle missing values, and provide probabilistic predictions across 21 quantiles. Mistral AI has also released the source code for Timestral, enabling users to run the models and reproduce benchmark results. AI
IMPACT Provides new foundation models for time series forecasting, potentially improving accuracy and efficiency in applications relying on temporal data.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →