# Mistral's Timestral brings forecasting to open weights

> The French lab releases a compact 209M-parameter foundation model aimed squarely at time-series prediction.

Published by The Open Weights on Oct 14, 2026. Canonical: https://theopenweights.com/news/timestral-209m-3lhq

## Key facts

- Company: Mistral AI
- Model: Timestral-209M
- Version: 1.0
- Category: Text / LLM
- Modalities: Text / LLM
- License: Other (Open weights, commercial use allowed)
- Parameters: 209M
- Precision: FP32
- Weights size: 0.8 GB
- Significance: notable
- Published: 2026-10-14
- Last verified: 2026-10-26
- Hugging Face: https://huggingface.co/mistralai/Timestral-209M
- Canonical URL: https://theopenweights.com/news/timestral-209m-3lhq

Mistral AI has published [Timestral-209M](https://huggingface.co/mistralai/Timestral-209M), a compact foundation model built for time-series forecasting rather than the conversational text generation the company is best known for. At roughly 209 million parameters, it sits well under the billion-parameter mark, making it light enough to run in constrained environments.

## A different kind of model

Forecasting foundation models are trained to predict future values from historical sequences — demand planning, energy loads, sensor readings, financial series and the like. By offering open weights in this category, Mistral is extending its portfolio beyond chat and code into a domain that has typically been served by bespoke statistical methods or narrowly trained neural networks.

Key details from the release:

- Around 209 million parameters, a deliberately small footprint
- Focused on time-series forecasting as its primary task
- Distributed on Hugging Face under a custom license

**Why it matters:** General-purpose forecasting models promise to cut the setup cost of building a predictor for every new dataset. A small, openly available option lowers the barrier for teams that want to experiment without committing to a proprietary API or training from scratch.

Mistral has not attached benchmark figures or a context window to the public record, so practitioners will want to validate Timestral against their own data before deploying it. Full model details are available on the [Hugging Face repository](https://huggingface.co/mistralai/Timestral-209M).

## Get the model

- [Hugging Face](https://huggingface.co/mistralai/Timestral-209M)

## Sources

- [mistralai/Timestral-209M](https://huggingface.co/mistralai/Timestral-209M) — Hugging Face, Oct 14, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "Mistral's Timestral brings forecasting to open weights", The Open Weights, Oct 14, 2026, https://theopenweights.com/news/timestral-209m-3lhq