What to provide
Enter 12 to 512 consecutive, evenly spaced observations and choose a horizon of 1 to 24 periods.
Example input
Series: 12, 15, 17, 16, 19, 18, 20, 24, 22, 25, 26, 29 Horizon: 6
How it works
Timestral-209M produces a forecast distribution. Studio displays P10, P50 and P90 and calculates the median’s mean absolute error over held-out periods.
How to read the result
Compare intervals and validation error with another model on the same data. Studio takes one series per run.
Credits and licenses
Timestral-209M · Mistral AI (2026)Modified MIT · Apache 2.0
The weights use a Modified MIT license. This permission is subject to a USD 20 million threshold on the prior month’s global consolidated revenue of the company or employer exercising the rights. The inference code uses Apache 2.0.