Mistral launches forecasting models that can skip the GPU
The four-model Timestral family forecasts related data streams and planned inputs; Mistral says its results have been submitted to GIFT-Eval and TIME.
By Ryan Merket · Published
Primary source: Mistral Newsroom
Why it matters
Timestral takes Mistral beyond language models into operational forecasting, where CPU inference and customer-controlled deployment could fit industrial systems. Its benchmark leadership, however, remains a company claim until the results and evaluation setup can be compared directly.
Mistral, the AI lab co-founded by Arthur Mensch, launched Timestral on October 15th, 2026, its first family of time-series foundation models, built to forecast operational data such as factory sensors, energy use and financial flows. The four models range from 5 million to 209 million parameters, and Mistral says they can run forecasts on standard server CPUs without a GPU.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.