With Timestral, Mistral Takes On Austria’s NXAI in Time-Series A.I.
Trending Topics Jakob Steinschaden ● Covered by 2 sources
Mistral launched Timestral, a new set of models for time-series forecasting. It puts the French AI firm right against Linz’s NXAI, including a company it already bought.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Mistral has stepped into time-series forecasting with Timestral, its first model family for predicting how data series change over time. The French company is aiming at the same industrial use cases NXAI has been chasing from Linz, from power demand and server load to sensor data and supply chains.
Timestral comes in four sizes: 5, 28, 123 and 209 million parameters. Mistral says the models can read up to 32,768 past data points and forecast as far as 1,024 steps ahead in one shot, even across several related series. That kind of reach is the pitch. So is speed. The company says the models run on ordinary server CPUs, not GPUs, and that a one-day forecast from three weeks of 15-minute data takes about 140 milliseconds with the largest model and about 10 milliseconds with the smallest.
Mistral is also pushing the usual open-ish distribution story. The weights are on Hugging Face under a modified MIT license, and businesses can fine-tune the models on their own data and deploy them in their own data centers. In Mistral’s own tables, the largest Timestral model ranks first among pretrained models on GIFT-Eval, TIME and fev-bench, ahead of Google’s TimesFM-3, Amazon’s Chronos-2 and Datadog’s Toto.
NXAI is not just a random rival here. Mistral acquired Emmi AI this year, and Emmi grew out of NXAI and Johannes Kepler University. That matters because Linz has become a Mistral industrial-AI hub, while NXAI is still selling TiRex and TiRex-2 to the same sort of buyers: carmakers, energy firms and other industrial customers. The irony is tidy. Mistral bought into the local ecosystem, then walked back in as competition.
The numbers suggest the field is moving fast. Mistral says its 28-million-parameter Timestral model already beats Chronos-2, while even the smaller Timestral models top TiRex-2 on the company’s own leaderboards. But the rankings are not locked in. Mistral’s results are self-reported, independent confirmation was still pending, and the gap over t0-gamma on GIFT-Eval is tiny.
My take — AI-written commentary, not fact-checked reporting
This is the kind of AI fight that actually matters: not chatbot theater, but who owns the boring industrial plumbing. Mistral is doing the sensible thing by chasing real deployment, and NXAI’s edge is staying close to the machine, where the GPU fan club can’t hear itself think. That's the market now: less hype, more control systems and cooler servers.
Read more about this at: Trending Topics