Timestral 1.0 123M
Timestral 123M provides the second-largest checkpoint in Mistral's forecasting family. It uses 20 transformer layers and a 640-dimensional internal representation. It retains the same published history length and single-pass forecast horizon as the larger 209M release.
Preview
Use numeric observations and related variables
The model accepts numeric or categorical covariates, including values known in advance. Categorical covariates use target encoding. Missing observations are marked with NaN, and each 32-step patch includes indicators for missing values. Forecasts include 21 quantiles.Timestral 123M official model card
Download the forecasting checkpoint
The official model page links to the 123M weights. It does not list a Mistral serverless forecasting identifier. The Modified MIT license needs a separate review from the standard MIT license.Timestral 1.0 123M model page
Compare forecasts on unseen periods
Practical guidance. The 123M size is a useful comparison point when many series must be processed within a fixed runtime budget. Measure the whole workload, including the number of variables and the length of history. A smaller parameter count alone does not specify throughput for that workload.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | Downloadable |
| Model identifiers | Timestral-123M |
| Release class | Public preview; downloadable checkpointTimestral 1.0 123M model page |
| Parameters | 123 million nominal parametersTimestral 123M official model card |
| History limit | 32,768 time steps, not language tokensTimestral 123M official model card |
| Single-pass horizon | 1,024 future time stepsTimestral 123M official model card |
| License | Modified MIT; company revenue condition appliesTimestral Modified MIT license |
Related models
- Timestral 1.0 209M
Mistral AI · Timestral
The 209M Timestral model forecasts numeric series with a 24-layer architecture and quantile outputs.
Launch date not confirmedPreviewDownloadable - Timestral 1.0 28M
Mistral AI · Timestral
A 12-layer, 28M Timestral model that produces numeric forecasts with quantile estimates.
Launch date not confirmedPreviewDownloadable - Timestral 1.0 5M
Mistral AI · Timestral
The smallest Timestral release uses eight layers and 5M parameters for numeric time-series forecasting.
Launch date not confirmedPreviewDownloadable