GitHub radar
Forecast numeric time series without a GPU using Timestral
Timestral is a family of time series forecasting models from Mistral AI. Its largest version, Timestral-209M, ranked first among pretrained models on three major benchmarks.
Timestral is a family of models from Mistral AI, a company that builds AI models. These models forecast numeric time series — data that changes over time. For example, sales or temperature.
When it helps
Useful if you need to forecast demand, revenue, server load, or customer churn. No GPU needed — it runs on an ordinary laptop. Not for text, code, or agentic tasks: it only handles numbers.
Pros
- The largest model (209M) ranked first among pretrained models on three benchmarks.
- Runs on an ordinary laptop — no GPU required.
- Lightweight: weighs between 21 and 837 MB.
- Outputs a range of likely values, not one number.
Cons
- This is Mistral's own benchmark — no independent verification yet.
- Handles only numbers: no text, code, or agentic tasks.
How to set it up — step by step
- 1Open your AI agent — Claude Code or Codex — in your project folder.
- 2Send the agent the text from the block below — it installs Timestral and runs a demo forecast.
- 3Wait while the agent downloads the model weights from Hugging Face — 21 to 837 MB.
- 4Check the result: the agent should show a forecast range on test data.
- 5For higher accuracy, ask the agent to run Timestral-209M — the largest, most accurate version.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Timestral from https://github.com/mistralai/timestral: run `pip install git+https://github.com/mistralai/timestral.git`, run the forecast example from the README on some test data, and show me the result.
Runs on a regular laptop — no GPU needed, model sizes range from 21 MB to 837 MB.
Open on GitHub▌ More finds