Mistral open-sources Timestral for multivariate time-series forecasting
A 5M-parameter model that tops zero-shot forecasting on GIFT-Eval means teams can run probabilistic forecasting on CPUs without GPU infrastructure, cutting deployment costs for demand or capacity planning. Engineers building forecasting pipelines should compare Timestral against TimesFM and Chronos on their own data. The 21-quantile output makes it practical for risk-aware decisions, not just point forecasts.
Mistral released Timestral, an open-source foundation model for multivariate probabilistic time-series forecasting. It predicts multiple variables jointly in one framework and supports known future external variables. Timestral took first place in zero-shot forecasting on benchmarks including GIFT-Eval and outputs 21 quantiles to quantify uncertainty. It handles long contexts of over 30,000 steps, and the smallest version has only 5M parameters, running directly on a regular CPU.
Mistral开源Timestral：高精度的多元时间序列预测
专为多元概率预测设计的基础模型，能在同一套框架下联合预测多个变量，并支持融入已知未来的外部变量。它在GIFT-Eval等多个主流公开测试集上拿下了零样本预测第一名，同时给出了21个预测分位数来量化未来走势的不确定性。模型不仅支持最长超过3万步的长期上下文，而且最小版本仅5M参数，普通CPU即可直接运行。
项目地址：https://t.co/mx7gxfjPu8
