Mistral AI is a French frontier-AI startup founded in 2023 by researchers from Meta and Google DeepMind. The early releases (Mistral 7B in 2023, Mixtral 8x7B in late 2023, Mixtral 8x22B in 2024) shipped under Apache 2.0 and reset the open-weights capability ceiling for several months. Mistral 7B in particular established the "small model that punches above its weight" pattern that subsequent open-weights releases iterated on (sliding-window attention, grouped-query attention). Mistral matters for two reasons. First, it is the credible European frontier-AI presence; sovereignty conversations in the EU specifically often cite Mistral as the existence proof that European AI does not have to depend on US labs. Second, the drift in their licensing posture is the canonical example of "open at the start, restrictive at the flagship" that other open-weights labs have followed. Mistral Large, their most capable model, is API-only and not open-weights. Compared to siblings: Llama (similar drift), Qwen (more durably open), DeepSeek (similar Mistral-style openness for the moment), OLMo (truly open across data and code). Production-readiness varies by tier. The Apache 2.0 small models are durable and widely deployed. The flagship API-only models compete with closed peers (Claude, GPT) but lock you into Mistral's roadmap if you depend on them. Strategic question for Mistral: does the European-sovereignty pitch sustain the company without the open releases that built its reputation in the first place?
The Stack · Weights · Open source
Mistral / Mixtral
French lab; older open releases under Apache 2.0; flagships increasingly API-only or under research-tier licenses.
Sources
- Mistral AI https://mistral.ai/
- Mistral 7B Paper https://arxiv.org/abs/2310.06825
- Mixtral of Experts (Mixtral 8x7B paper) https://arxiv.org/abs/2401.04088
- Mistral Models on HuggingFace https://huggingface.co/mistralai
- en.wikipedia.org (audit-verified) https://en.wikipedia.org/wiki/Mistral_AI
- mistral.ai (audit-verified) https://mistral.ai/models
Want a follow-up? Ask the chat about Mistral / Mixtral in context. It will compare to siblings at the same layer and ground every claim in the wiki.
Other projects at the Weights layer
9 siblings · ordered open first
- Qwen (Alibaba) Open source
Alibaba's aggressive open-weights series (Qwen 2.5 / 3); Apache 2.0 across most sizes; full-precision weights available.
- DeepSeek V3 / R1 Open source
Cost-quality reset; V3 papers documented architectural innovations (MoE, MLA, aux-loss-free MoE); R1 open reasoning model.
- OLMo (AI2) Open source
The only major model family meeting the strictest reading of OSAID: data (Dolma), training code, and weights all published.
- Phi (Microsoft) Open source
Small open models heavy on synthetic-data training; MIT license; cost-effective inference at edge sizes.
- Kimi (Moonshot AI) Open source
Chinese open-weights series; emphasis on long-context performance.
- GLM (Zhipu AI) Open source
Tsinghua-spinoff; ChatGLM and GLM-4 families; Apache 2.0 for major releases.
- Yi (01.AI) Open source
Kai-Fu Lee's Chinese open model family (Yi-34B etc.); Apache 2.0.
- Llama (Meta) Source available
Meta's open-weights family; dominant in usage; license carries a 700M-MAU clause and acceptable-use restrictions.
- Gemma (Google) Source available
Google's open-weights siblings to Gemini; source-available, not OSI-approved.