Mistral releases Timestral, a time-series forecasting model family
In short: Mistral AI has launched Timestral, its first family of time-series foundation models, available in four sizes from 5M to 209M parameters. The largest model ranks first on all three major public forecasting benchmarks (GIFT-Eval, TIME, fev-bench), beating previous state-of-the-art models that have more parameters. The models produce probabilistic, multivariate forecasts, support past and known-future covariates, and run efficiently on CPUs without needing a GPU.
This summary was generated automatically by AI from Mistral AI's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1New model family: 5M, 28M, 123M and 209M parameter versions, all sharing the same architecture
- 2209M model ranks #1 on GIFT-Eval, TIME and fev-bench benchmarks using fewer parameters than prior leaders
- 3Supports multivariate, probabilistic forecasts (21 quantiles) with past-only and known-future covariates
- 4Context length of 32,768 time steps
- 5Runs on standard CPUs: 10ms per forecast with the 5M model, 140ms with the 209M model on an 8-thread server
- 6Available for fine-tuning via Mistral Forge, deployable on-prem, VPC or air-gapped
- 7Released under a Modified MIT licence with a public GitHub repository
Why it matters
Time-series forecasting underpins operational decisions like capacity planning, equipment maintenance, energy and finance, and this release shows foundation-model-style zero-shot forecasting can now match or beat task-specific models while running cheaply on CPUs instead of GPUs.
Sources
- Mistral AIOfficialPrimary sourceOriginal article →„Introducing Timestral“15 Oct 2026, 15:00
- Published by source
- 15 Oct 2026, 15:00
- Found by our system
- 15 Oct 2026, 18:11
- Summary generated
- 15 Oct 2026, 18:14
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy