Loading the catalogue…
Loading the catalogue…
Mistral AI is a Paris-based, EU-native model creator positioned as Europe's sovereignty-focused answer to US and Chinese labs, with GDPR-by-default posture, SOC2/ISO27001 certification, and a signed EU GPAI Code of Practice commitment. However, its 'sovereign' story is complicated by heavy reliance on Microsoft Azure and Google Cloud for training/inference (including a Feb 2025 US-region expansion), a pending CNIL complaint over opt-out UX, and a licensing model that has drifted from fully open toward more proprietary/restrictive terms over time. Regulated EU customers get real jurisdictional advantages over US labs, but should verify subprocessor routing and licence terms per model.
Mistral's cloud subprocessor footprint (Azure, and Google Cloud including a US region added in 2025) can introduce indirect US CLOUD Act exposure at the infrastructure layer even though the Mistral entity itself is French, and its largest model (Mistral Large) has historically been distributed primarily via Microsoft Azure.
A pending (undecided) CNIL complaint alleges Mistral's free-tier training opt-out process was more cumbersome than for paid subscribers, a potential Article 12 GDPR transparency issue.
A former French government digital-affairs official (Cédric O) took an early advisory-founder equity stake in Mistral shortly after leaving public office, raising a 'revolving door' governance question relevant to independence from the French state.
Independent 2024 benchmark testing found Mistral's Mixtral model reproduced copyrighted book text verbatim in 22% of tested prompts, higher than Meta's LLaMA-2 or Anthropic's Claude 2 in the same test.
Stav’s assessment
Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.
Open-weights 119B MoE model from Mistral AI (128 experts, 4 active, 6.5B active per token) unifying instruct, opt-in reasoning, vision, and coding in a single 262K-context checkpoint with interactive-class latency.
Mistral AI's open-weight flagship: a 675B-parameter (41B active) Granular MoE model with native vision input, function calling, and a 262,144-token context window, built for long-document and multimodal workflows.
Mistral AI's closed-weight, 256K-context code specialist for fill-in-the-middle completion, code correction, and test generation.
Mistral signed the EU's voluntary GPAI Code of Practice in 2025, an early-adopter signal aligned with Article 53 transparency and copyright obligations under the AI Act.
Mistral holds SOC 2 Type II and ISO 27001/27701 certifications, published via its own Trust Center and Help Center.
Mistral operates a dedicated Legal Center documenting AI Act compliance materials per model, plus a public Trust Center for security documentation requests.
Continues to release genuinely open-weight models under Apache 2.0 (e.g. Mistral Large 3, Mistral Small, Mistral Nemo), supporting self-hosted deployment for customers needing full data control.
Backed by a large strategic investment and board presence from ASML, Europe's most valuable tech company, alongside continued enterprise traction (HSBC, CMA CGM, European Patent Office) and a French national civil-service deployment.
Published safeguards & certifications