Mistral released four downloadable Timestral weights with model sizes, forecasting inputs and public benchmark results.
Mistral releases four open-weight Timestral forecasting models
2026-10-15Verified announcementTRIAL
Mistral released four downloadable Timestral time-series models at 5M, 28M, 123M and 209M parameters. One checkpoint handles univariate and multivariate probabilistic forecasts with past and known-future covariates. The public 209M card lists 32,768 time steps of context, a 1,024-step forecast horizon and 21 quantiles. Its GIFT-Eval, TIME and fev-bench rankings are vendor-submitted and not independently reproduced.
Availability: downloadable Hugging Face weights; CPU and GPU inference
What this means for you
For engineers building capacity, demand or anomaly forecasts, Timestral provides a small unified baseline that can run on CPU or GPU. Open weights lower the trial barrier, but leaderboard claims and the modified MIT terms still need local review.
Evidence & related material
Technical details & scores
| Models | 5M, 28M, 123M, 209M parameters |
|---|---|
| Context | 32,768 time steps; 1,024-step forecast horizon on 209M |
| Output | 21 quantiles for probabilistic forecasting |
| Inputs | univariate or multivariate series; past and known-future covariates |
| Benchmarks | Mistral reports first place for 209M on GIFT-Eval, TIME and fev-bench; no independent reproduction in this run |
| Availability | downloadable Hugging Face weights; CPU and GPU inference |
| License | modified MIT with third-party-rights clause |
| Time Basis | official announcement provides date only; no hourly publication time |
Technical impact 3
Engineering value 4
Adoption signal 2
Maturity 3
Verification cost 2
Risk 2