News/Models
Try nowModels·IncrementalOfficialstableUpdated Oct 19·Event Jul 15, 2026·First seen Oct 19
Mistral Releases Timestral Time-Series Foundation Models
LaunchPractical
Try it
Something you can actually use or run today.
On July 15, 2026, Mistral AI released Timestral, a series of four models ranging from 5M to 209M parameters designed for multivariate time-series forecasting.
It offers a lightweight, CPU-efficient alternative to using general LLMs for numerical data, optimized for edge and industrial IoT environments.
AILookup take
Mistral is correctly identifying that time-series data is a weakness of general LLMs and is offering a hyper-specialized tool for it. The tiny parameter counts show a clear focus on the industrial edge market rather than the cloud.
Who cares
IoT engineersquantitative analystsedge computing developers
Watch next
Integration of Timestral into major industrial monitoring platforms.
Details
- Provides a specialized, lightweight alternative to repurposing large language models for numerical sequence data.
- CPU-first architecture and small parameter counts optimize performance for edge computing and real-time analytics.
- Integrated into the Mistral ecosystem with support for fine-tuning via Mistral Forge.
Consensus
All sources confirm the release, parameter ranges, and the focus on time-series forecasting.
Related articles (1)
Timestral: Mistral's time-series foundation modelsHacker News· 1 stories
More in Models
Anthropic Releases Claude 5.5 Series Including Opus, Sonnet, and Haiku23 sources · Oct 16Elorian Glimpse Launches Visual Reasoning Model Outperforming gpt-6 on Charts1 source · Oct 16Google AI Model Releases and Platform Initiatives12 sources · Oct 16Ibm Releases Granite 4.1 Switch for Chat and Decisions1 source · Oct 15NVIDIA Releases Nemotron 3.5 Super Vl and Bolt for Edge Agents1 source · Oct 15