Mistral Releases Timestral: 209M Time Series Model
Mistral AI has launched Timestral, a time series foundation model family ranging from 5M to 209M parameters, with the 209M variant outperforming Google's TimesFM-3 on zero-shot benchmarks and running on CPU.
Stock photo for illustration only, not from the actual event
- Mistral released Timestral, a time series foundation model family ranging from 5M to 209M parameters.
- The 209M model leads GIFT-Eval, TIME, and fev-bench among zero-shot models.
- The model family is engineered to run efficiently on standard CPUs.
- Reported by MarkTechPost on October 18, 2026.
Mistral AI has officially unveiled Timestral, a new family of time series foundation models offering parameter scales from 5 million (5M) up to 209 million (209M). This release aims to provide robust forecasting capabilities tailored for complex sequential data analysis across various industries.
A major highlight of this rollout is the 209M parameter model, which successfully outperforms Google's TimesFM-3 across several key evaluation benchmarks including GIFT-Eval, TIME, and fev-bench within the zero-shot model category, establishing a strong performance benchmark in temporal forecasting.
Stock photo for illustration only, not from the actual event
Furthermore, the Timestral model family is designed to execute efficiently on standard CPUs. This architectural advantage significantly lowers the hardware barriers and infrastructure costs typically associated with deploying heavy specialized accelerators for large-scale time series forecasting.
The ability of Mistral's Timestral models to run directly on CPUs marks a significant milestone in making advanced temporal forecasting accessible to smaller teams and enterprises without requiring expensive GPU clusters. Outperforming Google's TimesFM-3 highlights the intensifying competition in specialized foundation models designed for business analytics, financial forecasting, and operational planning.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment