Open Weight Models
Home›Compare›Europe deployment
Europe deployment · source-first decision page

Open-Weight Models for EU Self-Hosting

A deployment shortlist for EU/EEA self-hosting that keeps model provider origin separate from inference location, data residency and regulatory role.

Updated 1 Oct 20268 referenced modelsNo universal ranking
Direct answer

EU self-hosting is an infrastructure property, not a provider-nationality badge. A model from a US, Chinese or EU publisher can be deployed on EU/EEA infrastructure when the weights and software permit it. The residency analysis must cover inference, RAG, embeddings, logs, monitoring, backups and subprocessors; self-hosting alone does not establish GDPR or AI Act compliance.

01Choose an EU/EEA infrastructure boundary before selecting external services.
02Keep provider origin separate from where prompts and documents are processed.
03Review the exact checkpoint license and software stack.
04Document legal role, processors, retention and international transfers separately.

Models to evaluate

ModelArchitecture / sizeLicenseHardware classProvider
Ministral 3 8B Instruct 2512Mistral AI8B-class · vision-language edge modelApache 2.0≤8B · consumer/local🇫🇷 France
Devstral Small 2 24B Instruct 2512Mistral AI24B · coding agent modelApache 2.017–32B · high-memory workstation🇫🇷 France
Mistral Small 4 119B A6BMistral AI119B / 6.5B active · multimodal MoEApache 2.0Model-specific · large / specialized🇫🇷 France
Granite 4.2 8BIBM8B · 128K enterprise reasoningApache 2.0≤8B · consumer/local🇺🇸 United States
Qwen3 14BQwen / Alibaba14B · dense general-purpose modelApache 2.09–16B · workstation/local🇨🇳 China
gpt-oss-20bOpenAI20B · compact reasoningApache 2.017–32B · high-memory workstation🇺🇸 United States
Gemma 3 12B ITGoogle DeepMind12B · multimodalGemma Terms9–16B · workstation/local🇺🇸 United States
Phi-4Microsoft14B · denseMIT9–16B · workstation/local🇺🇸 United States
Shortlist, not ranking: these candidates span different capability and hardware classes. Remove incompatible models first, then benchmark the remainder on the exact workload.

Decision criteria

Choose an EU/EEA infrastructure boundary before selecting external services.

Keep provider origin separate from where prompts and documents are processed.

Review the exact checkpoint license and software stack.

Document legal role, processors, retention and international transfers separately.

Deployment reality

Validate the exact checkpoint, precision or quantization, runtime, context length and concurrency target. Weight memory alone does not capture KV cache, runtime workspaces, multimodal encoders or distributed-serving overhead.

OWM keeps license, provider origin and data residency separate. A provider-country label is provenance metadata; the deployer determines where inference and connected services run.

Primary model sources

Ministral 3 8B Instruct 2512 — official model source ↗Devstral Small 2 24B Instruct 2512 — official model source ↗Mistral Small 4 119B A6B — official model source ↗Granite 4.2 8B — official model source ↗Qwen3 14B — official model source ↗gpt-oss-20b — official model source ↗Gemma 3 12B IT — official model source ↗Phi-4 — official model source ↗OWM model-selection framework →