Timestral

Not yet independently verified. Every detail here comes from Mistral's own announcement of 15 October 2026 and its model cards; no independent report was found. The benchmark rankings are Mistral's own reading of public leaderboards, so no score is stored. Mistral states no hosted price. We will update this when it can be confirmed, and remove this note.

Timestral is Mistral AI's first family of time-series foundation models, released on 15 October 2026. It forecasts numerical series such as sensor readings, observability metrics, sales volumes or financial flows without being trained on the customer's own data first, a method known as zero-shot forecasting. The family comes in four sizes, 5 million, 28 million, 123 million and 209 million parameters, all sharing one architecture and training recipe.

What it does

Timestral produces probabilistic forecasts: instead of a single predicted value it returns 21 quantiles, from the 1st to the 99th percentile, so a team can see how wide the uncertainty is and set alert thresholds against it. It forecasts one series or several related series together, and it accepts covariates, both those known only in the past and those known in advance, such as a production schedule or a planned shutdown, including categorical ones like a machine's state. It reads up to 32,768 past time steps and forecasts up to 1,024 steps in one pass, extending further block by block. Mistral says it needs no GPU: on an eight-thread server CPU, a one-day forecast from three weeks of 15-minute data takes 10 milliseconds on the 5M model and 140 milliseconds on the 209M model.

Who it is for

Mistral aims it at industrial plants forecasting process sensors, finance teams forecasting volumes and volatility, and operations teams planning capacity, energy, supply chains and maintenance. It can run inside an existing monitoring pipeline or alongside Mistral's language models in agent workflows that investigate an anomaly once it is flagged. In its own evaluation Mistral reports that the 209M model ranks first among pre-trained zero-shot models on the GIFT-Eval, TIME and fev-bench benchmarks, ahead of TimesFM-3 and Chronos-2; these are the vendor's readings of public leaderboards and no score is stored here.

Key facts

Maker: Mistral AI Released: 15 October 2026 Sizes: 5M, 28M, 123M and 209M parameters Context: up to 32,768 time steps; forecast horizon up to 1,024 steps per pass Licence: Modified MIT, weights on Hugging Face and code on GitHub Enterprise route: Mistral Forge, for fine-tuning on proprietary data, deployable on the customer's own hardware, including air-gapped sites