What happened
Mistral AI introduced a new suite of language models, including the proprietary, API-accessible Mistral Large 3 (multimodal), Codestral (supporting over 80 programming languages), and Magistral Small/Medium for chain-of-thought tasks. Concurrently, Mistral released open-source models, which are downloadable and customisable, enabling on-device deployment for diverse applications and local data processing. This expands model availability for high-end and customisable use cases, independent of external infrastructure.
Why it matters
The availability of downloadable, customisable open-source models introduces a significant control gap for internal IT security and compliance teams. While enabling local data processing, this shifts the burden of model governance, security patching, and data lifecycle management from the vendor to the deploying organisation. This increases exposure to unmanaged model behaviours and raises due diligence requirements for platform operators regarding data privacy and intellectual property within custom deployments.
