[Late catch] Timestral: Mistral's first time-series foundation models, 5M to 209M, open weights
What is it?
Timestral is Mistral's first family of time-series foundation models. It forecasts sensor readings, demand or cash flow zero-shot, meaning without per-dataset training, and returns probabilistic forecasts over 21 quantiles. It comes in four sizes (5M, 28M, 123M, 209M parameters) with a 32,768-step context, and the weights are on GitHub and Hugging Face.
Why does it matter?
Mistral says the 209M model ranks first on GIFT-Eval, TIME and fev-benchVendor claim. It reports about 10 ms per forecast for the 5M model and about 140 ms for 209M on 8 CPU threadsVendor claim, so you can try demand forecasting or anomaly detection without a GPU. The announcement says modified MIT while the repo lists Apache-2.0, so check the license before commercial use.
Maturity
GACode and weights are public. The rankings are Mistral's own claim and the technical report has not been released yet.
How to track it
- Technical report: see whether the promised report discloses training data and evaluation setup.
- Leaderboards: check that Timestral appears on the official GIFT-Eval and TIME boards as claimed.
- License: watch for the announcement and repo license labels to be reconciled.