Community mirror of mistral.ai/news + official Hugging Face & GitHub.
25 updates · page 1 of 2
Mistral introduced Timestral, a probabilistic forecasting model in four sizes from 5M to 209M parameters that runs on CPU and is zero-shot ready.
Why it matters: Developers can forecast time series with a small CPU-friendly model without task-specific training.
Mistral published Timestral-123M on Hugging Face for time-series forecasting under a non-standard license.
Why it matters: Teams that want a mid-size Timestral model can use the 123M checkpoint.
Mistral published Timestral-28M on Hugging Face for time-series forecasting under a non-standard license.
Why it matters: Developers can use the 28M Timestral checkpoint when they want a small forecasting model.
Mistral published Timestral-5M on Hugging Face for time-series forecasting under a non-standard license.
Why it matters: Developers who need a tiny forecasting model can start with the 5M checkpoint.
Mistral published Timestral-209M on Hugging Face for time-series forecasting under a non-standard license.
Why it matters: Forecasters who need the largest Timestral size can download the 209M checkpoint.
Mistral Vibe v2.26.1 reports which process holds a stored session, adds Rust CLI /proxy-setup for proxy and SSL certificate settings, and has the narrator read each turn's summary aloud.
Why it matters: A client can show a session open elsewhere as read-only, and Rust CLI users can set proxy and SSL from /proxy-setup.
Mistral published LIDstral-Arabic, a text-classification model under the Apache-2.0 license.
Why it matters: Developers can use an Apache-2.0 model to classify Arabic text.
Mistral published Voxtral-Mini-4B-Realtime-Arabic, a model tagged for automatic speech recognition.
Why it matters: Speech apps that need Arabic can try this realtime Voxtral Mini model.
Mistral AI published an announcement introducing Mistral Large 4.
Why it matters: Developers watching Mistral can open the Mistral Large 4 post to see the new model.
Mistral Vibe v2.26.0 discovers built-in plugins shipped by an embedding app, warns when the selected model cannot view images, and lists background processes with live status and a Stop action.
Why it matters: Embedders can ship their own plugins, and users can stop a background process from the session panel.
Mistral is opening a Munich hub for Physics AI and Industrial AI research, partnering with German industry.
Why it matters: German industry partners get a local Mistral hub for physics and industrial AI research.
Mistral AI shipped Mistral Vibe v2.25.8 (patch release) on GitHub.
Why it matters: Small update (fixes/improvements) — generally safe to upgrade; skim the changelog for behavior changes.
Mistral AI shipped Mistral Vibe v2.25.7 (patch release) on GitHub.
Why it matters: Small update (fixes/improvements) — generally safe to upgrade; skim the changelog for behavior changes.
Mistral AI shipped Mistral Vibe v2.25.5 (patch release) on GitHub.
Why it matters: Small update (fixes/improvements) — generally safe to upgrade; skim the changelog for behavior changes.
Open, private and multilingual AI is coming to your web browser.
Why it matters: A new product capability you may be able to use right away, without code changes.