EU sovereign AI infrastructure
The European gateway to open AI models.
One OpenAI-compatible API to deploy powerful open models on EU sovereign infrastructure — with model provenance, licensing, security and jurisdiction verified by Frontière AI.
The best open-source models
One API, multiple providers, one visible European jurisdiction.
EU jurisdiction on sovereign models
Sovereign-first routing · zero retention
27
live models
28 total in catalog
14
EU sovereign models
outside US jurisdiction by design
11
Frontière Verified models
independently assessed
Qwen3 235B (instruct)EUQwen3.5 397B (multimodal)EUGLM-5.2 (Zhipu/Z.ai)EUDeepSeek V4 Flash 0731 (1M context)EULlama 3.3 70BEUQwen3.6 27B (multimodal)EUQwen3.5 9B (fast, budget)EUQwen3 32BEUQwen3 Coder 30BEUQwen2.5-VL 72B (vision)EU
GPT-OSS 120B (OpenAI)EUGPT-OSS 20B (OpenAI)EUMistral Small 3.2 24BEUKimi K3 (2.8T parameters, 1M context)USKimi K3 — EU sovereign variantEUMuse Spark 1.1 (1M context)USMuse Spark 1.2 (1M context)USGLM-5.3 Flash (Zhipu/Z.ai)USQwen3.8 Max (2.4T, 1M context)USQwen3.8 27B (multimodal)EUQwen3.8 Flash (Qwen)USDeepSeek R1 (671B MoE)USDeepSeek V4 Pro (345B MoE)USDeepSeek R1 (Distill Qwen 32B)USDeepSeek R1 (Distill Qwen 14B)USGLM-5.3 (Zhipu/Z.ai)USGemma 3 27B (Google)USDeepSeek V4.1 Flash (552B MoE, 1M context)USQwen3 235B (instruct)EUQwen3.5 397B (multimodal)EUGLM-5.2 (Zhipu/Z.ai)EUDeepSeek V4 Flash 0731 (1M context)EULlama 3.3 70BEUQwen3.6 27B (multimodal)EUQwen3.5 9B (fast, budget)EUQwen3 32BEUQwen3 Coder 30BEUQwen2.5-VL 72B (vision)EUGPT-OSS 120B (OpenAI)EUGPT-OSS 20B (OpenAI)EUMistral Small 3.2 24BEUKimi K3 (2.8T parameters, 1M context)USKimi K3 — EU sovereign variantEUMuse Spark 1.1 (1M context)USMuse Spark 1.2 (1M context)USGLM-5.3 Flash (Zhipu/Z.ai)USQwen3.8 Max (2.4T, 1M context)USQwen3.8 27B (multimodal)EUQwen3.8 Flash (Qwen)USDeepSeek R1 (671B MoE)USDeepSeek V4 Pro (345B MoE)USDeepSeek R1 (Distill Qwen 32B)USDeepSeek R1 (Distill Qwen 14B)USGLM-5.3 (Zhipu/Z.ai)USGemma 3 27B (Google)USDeepSeek V4.1 Flash (552B MoE, 1M context)USWhy Frontière AI
Frontier models, European rules
Sovereign, verifiable — when you choose it
Mature models run on managed clouds with no non-EU capital control (OVHcloud, Scaleway). Brand-new models, we deploy ourselves on dedicated EU servers — sovereign from day one. On EU sovereign infrastructure, prompts are designed to stay outside US jurisdiction.
OpenAI-compatible, two lines to change
Keep your existing SDK, scripts and tools: change the base_url and the key, that's it. One key unlocks the whole catalog — not one account per model vendor.
Prepaid, no subscription
Top up a balance starting at €10; every call draws it down. No commitment, no surprise invoice — when the balance hits zero, it stops. For company workloads and personal projects alike.
The Sovereign AI Gateway
The control layer between your enterprise and AI models
Frontière AI is not another model marketplace. It is the Sovereign AI Gateway: a control layer that sits between your teams and the models — deciding where each call runs, what each model is allowed to do, and what your compliance team can prove. One policy point, every model.
Every call passes through the control layer — one policy point, every model.
See how we compare →Frontière Verified
What exactly did we verify?
Every model is assessed before it goes live across six criteria — license, provenance, security, GDPR assessment, benchmarks, API compatibility. The checklist is public on every model page, and the same labels come back in the API response.
Built for AI agents
Your stack plugs in unchanged
n8n, LangChain, Vercel AI SDK, Open WebUI, LibreChat — they all speak the OpenAI protocol, so they all speak to Frontière AI. Measured function calling, per-model tool support disclosed, one key for the whole catalog.
Integration guides →GDPR, B2B & B2C
Everyone's data, protected the same way
One rule, whether you're a company or an individual: your data stays under EU jurisdiction on sovereign models — and when it doesn't, the label says so before you call.
Your company
A real answer for your DPO: sovereign models run on providers with no non-EU capital control — structured to stay outside US court-order reach. Verifiable per call via the API's sovereign field.
Your personal projects
Individuals get the exact same infrastructure, labels and protections as companies — prepaid from €10, no subscription, no account per model vendor. Your data deserves no less.
No ambiguity, ever
Every model carries “EU sovereign” or “Fast access (US)” on its card and in the API. How we draw the line: GDPR vs. the CLOUD Act →
Know where your AI data goes
Compare Frontière AI
The same comparison your procurement team would run: jurisdiction, verified models, one API, prepaid — against OpenAI, Anthropic and OpenRouter, with sources.
Catalog
Available models
Every model carries its hosting status — see also GET /api/v1/models (field sovereign).
Qwen3 235B (instruct)
Hosted via scaleway
Open weights·GPAIModel card ↗
Qwen3.5 397B (multimodal)
Hosted via scaleway
Open weights·GPAIModel card ↗
GLM-5.2 (Zhipu/Z.ai)
Hosted via scaleway
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
DeepSeek V4 Flash 0731 (1M context)
Hosted via scaleway
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
Llama 3.3 70B
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Qwen3.6 27B (multimodal)
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Qwen3.5 9B (fast, budget)
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Qwen3 32B
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Qwen3 Coder 30B
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Qwen2.5-VL 72B (vision)
Hosted via ovhcloud
Vision model — its provider rejects tool calls with a 400, so it cannot back an agent. Use it for image understanding, not for function calling.
Open weights·GPAIModel card ↗
GPT-OSS 120B (OpenAI)
Hosted via ovhcloud
Open weights·GPAIModel card ↗
GPT-OSS 20B (OpenAI)
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Mistral Small 3.2 24B
Hosted via ovhcloud
Open weights·GPAIModel card ↗
Kimi K3 (2.8T parameters, 1M context)
Hosted via modal
Reasoning model. First call after a quiet period can take a few minutes to warm up — then it responds in seconds.
Open weights·GPAIModel card ↗
Kimi K3 — EU sovereign variant
Self-hosted on dedicated EU servers — coming soon
Muse Spark 1.1 (1M context)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
Muse Spark 1.2 (1M context)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
GLM-5.3 Flash (Zhipu/Z.ai)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
Qwen3.8 Max (2.4T, 1M context)
Hosted via openrouter
Open weights·GPAIModel card ↗
Qwen3.8 27B (multimodal)
Hosted via ovhcloud
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
Qwen3.8 Flash (Qwen)
Hosted via openrouter
Open weights·GPAIModel card ↗
DeepSeek R1 (671B MoE)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
DeepSeek V4 Pro (345B MoE)
Hosted via openrouter
Open weights·GPAIModel card ↗
DeepSeek R1 (Distill Qwen 32B)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
DeepSeek R1 (Distill Qwen 14B)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
GLM-5.3 (Zhipu/Z.ai)
Hosted via openrouter
Reasoning model — thinks before answering. Omit max_tokens or allow at least ~1,000.
Open weights·GPAIModel card ↗
Gemma 3 27B (Google)
Hosted via openrouter
Open weights·GPAIModel card ↗
DeepSeek V4.1 Flash (552B MoE, 1M context)
Hosted via modal
Multimodal reasoning model (text + images). First call after a quiet period can take a while to warm up. Omit max_tokens or allow plenty of headroom for the chain of thought.
Open weights·GPAIModel card ↗
No non-EU capital or jurisdictional control
The label follows the SecNumCloud principle: it's not about where the servers are, it's about who controls the company operating them. OVHcloud, Scaleway, or our own dedicated EU servers — no US parent company, no US-licensed subsidiary: the structure keeps a US court order from reaching them directly.
27 models live today
Frontier models, day one — clearly labeled
Brand-new frontier models can be served first through Modal (a US company, even in its EU region) while we stand up the sovereign variant. It's always labeled “fast access”, never “sovereign” — you choose per call, with full information.
Read: GDPR vs. the CLOUD Act →How it works
Calling a model in minutes
Create an account
No password — a magic link by email, or Google. The account is created on first sign-in.
Top up from €10
Prepaid credit via Stripe. Every API call draws it down; at zero, it stops. No invoice surprises.
Change two lines
Point base_url at Frontière AI, paste your key — one key for the whole catalog, streaming included.
Enterprise
European AI infrastructure for enterprises
Dedicated EU deployments, volume pricing, DPA and audit support — for teams running AI where jurisdiction is not optional.
Your prompt's journey
Where your prompt actually goes
Before you send anything, Frontière AI shows you the destination — that's the whole point of the label.
Runs on OVHcloud, Scaleway or our own dedicated EU servers — providers with no non-EU capital control, designed to stay outside US jurisdiction.
OVHcloud · Scaleway · EU servers
✓ Designed to stay under EU jurisdiction
A brand-new frontier model served first through Modal (US-owned, even in its EU region) while we stand up the sovereign variant — always labeled before you call.
Modal (US)
✗ Subject to the US CLOUD Act
Labels are decided per model, shown on the card and in the API response — never hidden.
FAQ
The questions we actually get asked
Is Frontière AI GDPR-compliant?+
Yes on every model labeled EU sovereign: those run on providers with no non-EU capital control — OVHcloud, Scaleway, or our own dedicated EU servers — so the corporate structure keeps a US court order from reaching your prompts directly. Models we can only serve through US-controlled infrastructure carry the “fast access” label, on the card and in the API response, before you send anything.
What does “EU sovereign” mean exactly?+
It follows the SecNumCloud principle: what matters is not where the servers sit, but who controls the company operating them. A US provider with an EU region is still subject to the CLOUD Act, so it never earns the sovereign label here. That single rule is why there are two explicit tiers instead of one vague “hosted in Europe” claim.
Does Frontière AI keep my prompts?+
Not for API calls. The only things written to our database are what billing requires: which model, how many input and output tokens, and the amount charged — never the prompt or the reply. The web console is the single, deliberate exception: conversations stay in your browser unless you choose to sync a thread to your account. The provider serving the call does see the content — which is precisely why every model carries its jurisdiction label.
How much does it cost, and is there a subscription?+
No subscription, and no card kept on file. You top up a prepaid balance from €10 and every call draws it down, billed per token — from €0.0056 per 100,000 input tokens on the cheapest sovereign model. When the balance reaches zero, calls stop and nothing else happens.
Do I have to rewrite my code?+
No. The API is OpenAI-compatible: point base_url at https://getfrontiereai.eu/api/v1, swap the key, and your SDK, scripts and tools keep working — streaming included. One key covers the whole catalog instead of one account per model vendor.
Which models can I call?+
A curated selection — 27 live today out of 28 in the catalog, from Qwen3.5 397B and GLM-5.2 to Llama 3.3 70B, Mistral Small 3.2 and Kimi K3. Not every open-source model on the market: an entry is added only once we have verified who serves it, under which licence, and at what real cost.
Your data stays under EU jurisdiction. Starting today.
Create an account, top up €10, change two lines. GDPR-compliant from the first call — for company workloads and personal projects alike.
Stay in the loop
New models, open-source news, and tips — once a week, no spam.