Timestral
Summary
Timestral is a family of four time series foundation models (5M to 209M parameters) achieving state-of-the-art probabilistic forecasting on GIFT-Eval, TIME, and fev-bench via decoupled horizon forecasting, native future covariate support, and synthetic pretraining followed by mixed post-training.
View Cached Full Text
Cached at: 10/28/26, 05:25 AM
Paper page - Timestral
Source: https://huggingface.co/papers/2610.37585 Authors:
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
Abstract
Timestral is a family of time series foundation models that achieves state-of-the-art probabilistic forecasting through decoupled horizon forecasting, native future covariate support, and synthetic pretraining followed by mixed post-training.
We introduce Timestral, a family of fourtime series foundation modelsforprobabilistic 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 supportsknown-future covariateswithin a causal architecture; and a two-stage pipeline that pretrains exclusively on a diverse, large-scale synthetic corpus beforepost-trainingon 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.
View arXiv pageView PDFAdd to collection
Get this paper in your agent:
hf papers read 2610\.37585
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2610.37585 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2610.37585 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.37585 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Timestral: Mistral's time-series foundation models
Mistral 发布 Timestral 时间序列基础模型家族，包含 5M 到 209M 参数四个版本，其 209M 模型在 GIFT-Eval、TIME 和 fev-bench 三大基准上均排名第一，支持零样本概率多变量预测且可在 CPU 上推理。
mistralai/Timestral
Mistral 发布 Timestral 时间序列基础模型家族（5M/28M/123M/209M 参数），支持零样本预测、单变量与多变量预测及概率输出，209M 版本代码与权重已在 GitHub 和 Hugging Face 开源。
mistralai/Timestral-209M
Mistral AI 发布 Timestral 时间序列基础模型系列（5M/28M/123M/209M），209M 版本在 GIFT-Eval、TIME 和 fev-bench 上实现零样本预测的最先进精度，并开源代码。
How does a 102M-parameter transformer forecast multivariate time series?
This article provides a visual walkthrough of t0-alpha, a 101.6M-parameter foundation model for multivariate time-series forecasting that separates time attention from cross-variable group attention, achieving competitive CRPS scores on GIFT-Eval compared to larger models like TimesFM 2.5 and Chronos-2.
TimesFM-3: A zero-shot foundation model for multivariate forecasting
Google introduces TimesFM-3, a state-of-the-art zero-shot foundation model for multivariate time series forecasting, capable of handling multiple targets and covariates in a single forward pass without fine-tuning.