Mistral launches Timestral, a new time-series foundation model family
Discovery source: Mistral · Source date:
On October 15, 2026, Mistral announced Timestral, its first family of time-series foundation models. Mistral says the largest model has 209 million parameters, with four sizes ranging from 5M to 209M, and that the family supports probabilistic multivariate forecasting with past-only and known-future covariates. In Mistral’s benchmarks, Timestral ranks first on the public GIFT-Eval, TIME, and fev-bench tests, but those results are the company’s own reported measurements rather than independent evaluation. Mistral also says the models run efficiently on CPUs, do not require a GPU at inference, and are available for on-prem, VPC, or air-gapped deployment.
Check the sources
This is a NextGen-AI briefing, not a reproduction of the discovery source. A newsletter may cover several developments; consult the evidence below for this briefing’s specific subject. The source date refers to the discovery entry, not necessarily the announcement date.