Home/Latest News/Mistral News
Mistral News
Mistral news from AiLookout: the latest Mistral AI model releases, open weights, Le Chat features, funding and European AI policy, with sources linked.
Newest first, each with its sources linked.
About our Mistral coverage
Mistral AI is a Paris-based company that builds language models, many released with open weights, along with the Le Chat assistant and an API for developers. This hub collects AiLookout’s Mistral news, newest first: model releases and licenses, coding and reasoning models, Le Chat features, funding and partnerships, and the European AI policy debates in which the company plays a part.
Mistral’s own benchmark comparisons are reported as Mistral’s claims and linked to the release post or model card. Independent tests and third-party reporting are linked separately so readers can see how the claims hold up.
Use this page to follow Mistral news and the European side of the AI industry. For other open-weight model makers, see our DeepSeek news hub, and for the largest US labs, see OpenAI news and Claude news. Every story here links its primary sources, shows its publication date, and discloses the AI assistance used to write it. New stories are added as they are published, and older coverage stays available further down the list.
Latest Mistral News

Mistral releases Le Chonk open weights as the best open-weight model for coding
Mistral released the open weights of Mistral Large 4 (Le Chonk), its 1-trillion-parameter multimodal flagship, on 30 October, following a public preview earlier this month — with three additional weeks of reinforcement learning making it the best open-weight model for coding and improving reasoning efficiency by 30% fewer output tokens.

Mistral launches Voxtral 3 speech models and a voice agent SDK
Mistral's Voxtral 3 adds transcription for 50 languages, an expressive text-to-speech model and a LiveKit-based SDK for phone and web voice agents, priced from $0.0035 a minute.

Mistral's Arthur Mensch tells Europe to build frontier AI or surrender
In an essay posted on 16 October, Mistral AI's chief executive argues that leaving frontier models to the US and China would lock Europe into dependence, says open weights alone are not enough, and commits Mistral to six times more compute within a year.

CERN and Mistral plan a Large Physics Model trained on simulated LHC collisions
CERN and Mistral have signed a statement of intent to build a single foundation model for collider physics, trained on billions of simulated proton-proton collisions at the Large Hadron Collider. No budget, timeline or model size was disclosed.

Mistral releases Timestral, open time-series forecasting models that run on a CPU
Timestral is Mistral's first family of time-series foundation models: four sizes from 5M to 209M parameters that produce probabilistic, multivariate forecasts with covariates and run without a GPU. Mistral says the largest tops three public benchmarks; the weights carry a revenue cap for big companies.

Mistral and Morocco release open Voxtral speech-to-text for Arabic dialects and Darija
Voxtral Mini 4B Realtime Arabic, an Apache 2.0 model built with Morocco’s digital ministry, transcribes Modern Standard Arabic and 15 dialects as people speak, including Arabic-French-English code-switching. A companion classifier, LIDstral Arabic, tells Arabic dialects and other Arabic-script languages apart.

ESA launches Earth Virtual Expert, an open AI model for Earth observation questions
The European Space Agency opened EVE, a question-and-answer assistant for Earth observation and Earth science built with Pi School, Mistral AI and Imperative Space, and published its 24B model, code, datasets and benchmarks under open licences.

Mistral previews Le Chonk with unusually strong cyber benchmarks
Mistral Large 4 is a one-trillion-parameter multimodal model in API preview, with open weights promised by month-end and company-reported cyber results that need careful interpretation.

