Timestral
State-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready.
Are you the maker of Timestral? Claim this listing
What is Timestral?
Timestral is State-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready. It's designed for data teams forecasting sensor, demand or ops metrics who want a small zero-shot model.
Timestral (Oct 2026)
Community video
About Timestral
Key Strengths
- Open weights in four sizes (5M, 28M, 123M and 209M parameters), with Apache-2.0 inference code on GitHub
- Runs on CPU: Mistral reports 10 ms (5M) to 140 ms (209M) to forecast a day ahead from three weeks of 15-minute data on an 8-thread server
- Multivariate, probabilistic forecasts (21 quantiles) with past-only and known-future covariates and up to 32,768 steps of context
Limitations to Consider
- The #1 rankings on GIFT-Eval, TIME and fev-bench are Mistral's launch claims; the full technical report hasn't been published yet
- The weights use a 'Modified MIT' license, not plain MIT or Apache, so read the terms before commercial use
- A developer model in Public Preview, not an app: you run it in Python yourself, and fine-tuning goes through Mistral Forge
Best For
Data teams forecasting sensor, demand or ops metrics who want a small zero-shot model.
Skip it if: You want a point-and-click forecasting app.
Pricing
flexible pricing options
Free open weights on Hugging Face (Modified MIT license); inference code is Apache 2.0. Mistral Forge fine-tuning pricing is not published. (checked Oct 19, 2026)
Sources
- https://mistral.ai/news/timestral/
- https://github.com/mistralai/timestral
- https://huggingface.co/collections/mistralai/timestral
- https://docs.mistral.ai/models/timestral-1-0
Pros, cons and pricing compiled from public sources (official pricing/docs pages, published reviews, user reports), not hands-on testing.
Compare Timestral
Similar Tools in AI Art & Drawing
Sensia
Private creative studio — save one character identity, then change everything else: scene, outfit, lighting. Keep the same consistent character across every generation.
Try SensiaQwen-Image-2.1
This app lets you create new pictures just by describing them, or modify up to 10 uploaded images using a written prompt. You can also add a negative prompt, choose custom dimensions, and enable au...
Try Qwen-Image-2.1Burning Bridges AI
Burning Bridges AI is an AI video generator for making the viral Burning Bridges party-moment clip. Upload a photo, pick a scene, edit one of four ready-made party prompt recipes, compare model settings and preview 1080p demo videos in three styles.
Try Burning Bridges AIFrequently Asked Questions
What is Timestral?
Timestral is State-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready. It's designed for data teams forecasting sensor, demand or ops metrics who want a small zero-shot model.
How much does Timestral cost?
Timestral offers flexible pricing options You can check pricing on their official website.
Who should use Timestral?
Timestral is best suited to data teams forecasting sensor, demand or ops metrics who want a small zero-shot model — particularly users who need high-quality visual content creation.