
Volume 344, Issue 1 · 6 articles from 5 issues
Mistral releases weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter multimodal open-weight model for coding and defensive security.
Mistral Large 4 was released in public preview earlier in the month and the improved model's weights are now open.
It is a natively multimodal model with 1 trillion parameters, trained from scratch on European infrastructure, with three reasoning modes balancing cost and performance.
After three more weeks of reinforcement learning, Mistral says it leads open-weight models on FrontierCode and Terminal-Bench, matches the best on the Artificial Analysis Coding Agent Index with greater token efficiency, and improved in defensive cybersecurity.
Mistral pitches it for sovereign AI and self-hosted deployment.
Mistral AI launches a public preview of Mistral Large 4, a 1-trillion-parameter multimodal open-weight model with weights due this month.
Mistral AI launched a public preview of Mistral Large 4 (ML4), a natively multimodal model with 1 trillion parameters and 52 billion active, available through the Mistral Studio preview API with open weights due by the end of the month.
It was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters, and Mistral calls it state of the art among open models for cybersecurity, finance and law, citing 82% on an Artificial Analysis Cyber Index test and 93% on Cybench.
The model is being red-teamed with cybersecurity leaders, partners and state authorities before the weights are released.
Mistral released Large 4, an open-weight mixture-of-experts model with about 1.05 trillion parameters, aimed at coding and cybersecurity.
Mistral Large 4, nicknamed Le Chonk, is in public preview and has roughly 1.05 trillion parameters with 52 billion active, a 1 million token context window, multimodal input and more than 160 languages.
It scored 49.8 on the Artificial Analysis Coding Agent Index and placed second of five in a Surge AI blind coding test, behind Claude Opus 5.
Mistral says it scored 82% on a vulnerability reproduce-and-patch test, solved 93% of Cybench challenges and blocked 93.3% of attacks in Lakera's B3 benchmark.
It was trained on Mistral's own European NVIDIA Grace Blackwell GPUs after a EUR 3 billion raise in September 2026.
Mistral released Timestral, its first family of time-series forecasting foundation models, from 5M to 209M parameters.
Mistral released Timestral, its first family of time-series foundation models, in four sizes from 5M to 209M parameters, delivering probabilistic, multivariate forecasts with support for past and known-future covariates.
Mistral says the 209M model ranks first on GIFT-Eval, TIME and fev-bench with two-thirds the parameters of the previous state of the art.
The models run efficiently on CPUs, perform well zero-shot and can be fine-tuned through Mistral Forge.
Mistral AI launches Voxtral 3, a voice agent SDK and speech models for real-time conversations in more than 50 languages.
Mistral AI has introduced Voxtral 3, a voice agent SDK and speech models for real-time, multi-step conversations in more than 50 languages.
It combines transcription, natural speech synthesis with emotional adaptability and deep tool integration.
Voxtral 3 is available through Mistral's EU-hosted API or self-deployment, which Mistral says supports data sovereignty.
Target uses include contact centers, virtual assistants and hands-free copilots.
Mistral launched Mistral Large 4 (Le Chonk) and expanded Mistral Studio with managed agents, apps, workflows and observability.
Mistral AI positions Mistral Large 4, nicknamed Le Chonk, as an open model strong in coding, defensive cybersecurity and reasoning over complex documents.
Mistral Studio, its enterprise AI platform, lets teams build agents, workflows and apps connected to enterprise data and deploy them in the cloud, a VPC, on-premises or a mix.
Studio now adds managed agents and apps along with new capabilities for workflows, enterprise context, agent observability and evaluations.
Customers and partners already using Studio in production include ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services.