analiza aiqbench
This paper presents a family of time series foundation models (5M to 209M parameters) for probabilistic forecasting, with a new training design (Decoupled Horizon Forecasting) and a pretraining pipeline built mainly on synthetic data. The benchmark claims are strong, but they rest on three public evaluation suites, so readers should check how they hold up on their own forecasting workloads. Open weights and inference code make independent verification practical.
Timestral
arXiv:2610.37585v1 Announce Type: new Abstract: We introduce Timestral, a family of four time series foundation models for probabilistic forecasting, with 5M, 28M, 123M, and 209M parameters. Timestral sets a new state of the art on GIFT-Eval, TIME, and fev-bench, where Timestral-209M and Timestral-123M take the top two places on every metric. We achieve this with three ingredients: Decoupled Horizon Forecasting (DHF), which separates prediction tokens from the causally encoded context, so that longer horizons are trained without masking context and each observed value is both a forecast target and context for later forecasts; a covariate framework that natively supports known-future covariates within a causal architecture; and a two-stage pipeline that pretrains exclusively on a diverse, large-scale synthetic corpus before post-training on a mix of real and synthetic data with extended context. This report describes the model design and its benchmark results. Weights for all four models are available on Hugging Face and inference code on GitHub.
Abstract:We introduce Timestral, a family of four time series foundation models for probabilistic forecasting, with 5M, 28M, 123M, and 209M parameters. Timestral sets a new state of the art on GIFT-Eval, TIME, and fev-bench, where Timestral-209M and Timestral-123M take the top two places on every metric. We achieve this with three ingredients: Decoupled Horizon Forecasting (DHF), which separates prediction tokens from the causally encoded context, so that longer horizons are trained without masking context and each observed value is both a forecast target and context for later forecasts; a covariate framework that natively supports known-future covariates within a causal architecture; and a two-stage pipeline that pretrains exclusively on a diverse, large-scale synthetic corpus before post-training on a mix of real and synthetic data with extended context. This report describes the model design and its benchmark results. Weights for all four models are available on Hugging Face and inference code on GitHub.
Artificial Intelligence (cs.AI)
arXiv:2610.37585 [cs.AI]
(or arXiv:2610.37585v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.37585 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 245 cuvinte
Subiecte conexe
Introducing Timestral
Mistral AI · 15 oct. 2026
Zero-shot time-series foundation models against the Khepri v1.5 baselines for 96-hour carbon-intensity forecasts in nine Nordic bidding zones: a preregistered comparison
arXiv · 3 nov. 2026
Mistral AI Releases Timestral: A 209M Forecasting Model Family That Outperforms Google’s TimesFM-3
MarkTechPost · 19 oct. 2026
World Models, Trees and Foundation Models for Retail Forecasting: Evidence from Four Public Data Sets
arXiv · 21 oct. 2026