# Frontière AI

> EU sovereign OpenAI-compatible AI gateway giving businesses and individuals
> access to the most powerful open-source models (DeepSeek, Qwen, GLM, Llama,
> Mistral, Kimi, Muse Spark). Prepaid credit (minimum EUR 10), no subscription,
> prepaid model only. Every model is labeled EU sovereign or not — see /api/v1/models.

> Key differentiators: EU data residency, zero prompt retention, AI Act-ready
> positioning, GDPR-first architecture, sovereign providers (OVHcloud, Scaleway,
> self-hosted) alongside fast-access non-sovereign ones (Modal, OpenRouter).

## Quickstart
- [Home](https://getfrontiereai.eu): positioning, model catalog, pricing model
- [API documentation](https://getfrontiereai.eu/docs): authentication, chat completions, streaming, errors, prepaid billing
- [Model catalog page](https://getfrontiereai.eu/models): every model with its publisher, licence, size, hosting jurisdiction and real per-token price
- [Benchmarks](https://getfrontiereai.eu/benchmarks): Frontière Verified scores — 6-criteria independent verification for every model
- [Open Source AI Benchmark](https://getfrontiereai.eu/open-source-ai-benchmark): Independent model comparison with real benchmark scores — MMLU, GPQA, LiveCodeBench, Arena Elo, and EU sovereign pricing
- [Rankings](https://getfrontiereai.eu/rankings): model quality rankings — sorted by verified score, sovereignty, and price efficiency
- [Compare](https://getfrontiereai.eu/compare): side-by-side model comparison — benchmarks, pricing, context, tool calling, sovereignty
- [Pricing page](https://getfrontiereai.eu/pricing): per-model pricing, margin included, displayed per 100k tokens
- [Sign up](https://getfrontiereai.eu/login): create an account with magic link (email) or Google
- [API keys](https://getfrontiereai.eu/dashboard/keys): create an API key (after login)
- [Web console](https://getfrontiereai.eu/dashboard/chat): test models in your browser before using the API
- [Public preview](https://getfrontiereai.eu/preview): explore the platform without an account

## API Reference
- [GET /api/v1/models](https://getfrontiereai.eu/api/v1/models): list live models with sovereign, host, jurisdiction, pricing, compliance and verified fields (add ?include=all for coming-soon)
- [POST /api/v1/chat/completions](https://getfrontiereai.eu/docs): OpenAI-compatible chat completions with streaming (SSE), tool calling, and per-model pricing
- [POST /api/v1/embeddings](https://getfrontiereai.eu/embeddings): OpenAI-compatible embeddings for RAG and semantic search
- [MCP server /api/mcp](https://getfrontiereai.eu/mcp): Model Context Protocol endpoint for AI agents — 8 tools: model catalog, sovereign inference, embeddings, balance, and EU web search/fetch (stateless JSON-RPC 2.0, catalog tools need no key)
- [Authentication](https://getfrontiereai.eu/docs): Bearer API key in the Authorization header (keys start with sk-front-)
- [Error codes](https://getfrontiereai.eu/docs): OpenAI-compatible error format (400, 401, 402 insufficient credit, 429, 503)
- [Streaming](https://getfrontiereai.eu/docs): SSE with stream_options include_usage true for accurate per-request billing

## Models
- [Model catalog](https://getfrontiereai.eu/models): filterable table of all live and coming-soon models by jurisdiction tier
- [Model pricing](https://getfrontiereai.eu/pricing): per-model pricing with margin included, per 100k tokens
- [Embeddings models](https://getfrontiereai.eu/embeddings): vector embedding models for RAG, with measured dimensions and context
- [MCP for agents](https://getfrontiereai.eu/mcp): connect Claude Code, Cursor or any MCP client to EU-sovereign models — one URL, 8 tools

- [qwen3-235b](https://getfrontiereai.eu/models/qwen3-235b): Qwen · ? context · EU sovereign
- [qwen3.5-397b](https://getfrontiereai.eu/models/qwen3.5-397b): Qwen · ? context · EU sovereign
- [glm-5.2](https://getfrontiereai.eu/models/glm-5.2): Zhipu / Z.ai · ? context · EU sovereign
- [deepseek-v4-flash-0731](https://getfrontiereai.eu/models/deepseek-v4-flash-0731): DeepSeek · 1M context · EU sovereign
- [llama-3.3-70b](https://getfrontiereai.eu/models/llama-3.3-70b): Meta · ? context · EU sovereign
- [qwen3.6-27b](https://getfrontiereai.eu/models/qwen3.6-27b): Qwen · ? context · EU sovereign
- [qwen3.5-9b](https://getfrontiereai.eu/models/qwen3.5-9b): Qwen · ? context · EU sovereign
- [qwen3-32b](https://getfrontiereai.eu/models/qwen3-32b): Qwen · ? context · EU sovereign
- [qwen3-coder-30b](https://getfrontiereai.eu/models/qwen3-coder-30b): Qwen · ? context · EU sovereign
- [qwen2.5-vl-72b](https://getfrontiereai.eu/models/qwen2.5-vl-72b): Qwen · ? context · EU sovereign
- [gpt-oss-120b](https://getfrontiereai.eu/models/gpt-oss-120b): OpenAI · ? context · EU sovereign
- [gpt-oss-20b](https://getfrontiereai.eu/models/gpt-oss-20b): OpenAI · ? context · EU sovereign
- [mistral-small-3.2-24b](https://getfrontiereai.eu/models/mistral-small-3.2-24b): Mistral AI (France) · ? context · EU sovereign
- [kimi-k3](https://getfrontiereai.eu/models/kimi-k3): Moonshot AI · 1M context · fast access (non-sovereign)
- [muse-spark-1.1](https://getfrontiereai.eu/models/muse-spark-1.1): Meta · 1M context · fast access (non-sovereign)
- [muse-spark-1.2](https://getfrontiereai.eu/models/muse-spark-1.2): Meta · 1M context · fast access (non-sovereign)
- [glm-5.3-flash](https://getfrontiereai.eu/models/glm-5.3-flash): Zhipu / Z.ai · 1.3M context · fast access (non-sovereign)
- [qwen3.8-max](https://getfrontiereai.eu/models/qwen3.8-max): Qwen · 1M context · fast access (non-sovereign)
- [qwen3.8-27b](https://getfrontiereai.eu/models/qwen3.8-27b): Qwen · 256k context · EU sovereign
- [qwen3.8-flash](https://getfrontiereai.eu/models/qwen3.8-flash): Qwen · 128k context · fast access (non-sovereign)
- [deepseek-r1](https://getfrontiereai.eu/models/deepseek-r1): DeepSeek · 128k context · fast access (non-sovereign)
- [deepseek-v4-pro](https://getfrontiereai.eu/models/deepseek-v4-pro): DeepSeek · 128k context · fast access (non-sovereign)
- [deepseek-r1-distill-qwen-32b](https://getfrontiereai.eu/models/deepseek-r1-distill-qwen-32b): DeepSeek · 128k context · fast access (non-sovereign)
- [deepseek-r1-distill-qwen-14b](https://getfrontiereai.eu/models/deepseek-r1-distill-qwen-14b): DeepSeek · 128k context · fast access (non-sovereign)
- [glm-5.3](https://getfrontiereai.eu/models/glm-5.3): Zhipu / Z.ai · 1.3M context · fast access (non-sovereign)
- [gemma-3-27b](https://getfrontiereai.eu/models/gemma-3-27b): Google · 128k context · fast access (non-sovereign)
- [deepseek-v4.1-flash](https://getfrontiereai.eu/models/deepseek-v4.1-flash): DeepSeek · 1M context · fast access (non-sovereign)

### Embedding models
- [bge-m3](https://getfrontiereai.eu/embeddings): BAAI · 1024 dimensions
- [bge-multilingual-gemma2](https://getfrontiereai.eu/embeddings): BAAI · 3584 dimensions
- [qwen3-embedding-8b](https://getfrontiereai.eu/embeddings): Qwen · 4096 dimensions

## Sovereign AI and Compliance
- [Why Frontière AI](https://getfrontiereai.eu/why-frontiere): the five-pillar positioning — European jurisdiction, one API, verified models, no lock-in, transparent routing, plus an enterprise use-case map and a comparison with OpenAI, Anthropic, OpenRouter
- [Enterprise](https://getfrontiereai.eu/enterprise): enterprise arrangements — private deployments and dedicated EU infrastructure (on request), volume pricing, DPA under GDPR Article 28, contractual SLA on dedicated arrangements, audit support; what enterprise-compliant AI requires in practice
- [How we determine sovereignty](https://getfrontiereai.eu/sovereignty): the four questions behind a model's label, the SecNumCloud principle, and the deterministic Sovereignty Score formula (5 weighted factors, 0-100)
- [AI Act readiness](https://getfrontiereai.eu/ai-act): how Frontière AI positions itself for the EU AI Act — sovereign infrastructure, data retention, sub-processors, licensing
- [GDPR vs CLOUD Act (blog)](https://getfrontiereai.eu/blog/gdpr-vs-cloud-act): why hosted in EU does not equal sovereign — the jurisdiction matters more than the data center location
- [Sovereign vs non-sovereign providers](https://getfrontiereai.eu/models): EU sovereign (OVHcloud, Scaleway, self-hosted) vs fast access (Modal, OpenRouter) — every model is labeled
- [Zero data retention](https://getfrontiereai.eu/docs): prompts are not stored or used for training
- [Frontière Verified](https://getfrontiereai.eu/models): per-model verification checklist — 6 criteria (license, provenance, security, GDPR, benchmarked, API available)

## Features
- EU sovereign hosting — models on OVHcloud AI Endpoints, Scaleway Managed Inference, or self-hosted servers in Europe
- Prepaid credit — no subscription, no surprise invoices. Top up minimum EUR 10, calls stop at zero balance
- OpenAI-compatible API — drop-in replacement: change the base URL and API key, everything else works
- Streaming (SSE) — real-time token streaming with usage included in the final chunk
- Tool calling — measured per model (see /api/v1/models for tool_calling field)
- AI Act-ready — sovereign infrastructure, no non-EU capital control, zero prompt retention
- Per-model compliance data — open weights, license, GPAI status, systemic risk, model card, training docs
- Frontière Verified — 6-criteria per-model verification (license verified, provenance verified, security analyzed, GDPR assessed, benchmarked, API available)
- Reasoning models — support for models with chain-of-thought (GLM-5.2)
- Multimodal — vision-language models (Qwen2.5-VL-72B)
- Cache-aware billing — cache read tokens billed at the lower provider rate times 1.4 margin

## Integrations
- [Integrations hub](https://getfrontiereai.eu/integrations): running AI agents on EU-hosted models — the three settings, tool-calling support measured per model
- [n8n integration](https://getfrontiereai.eu/integrations/n8n): Point the OpenAI credential at Frontière AI and the AI Agent node runs on EU-hosted models — including the tool calls the agent depends on.
- [Microsoft Foundry integration](https://getfrontiereai.eu/integrations/azure-foundry): A bring-your-own-model connection makes every catalog model selectable in Foundry Agent Service, with inference staying in Europe.

### Verified settings (OpenAI-compatible platforms)
- [LangChain (Python)](https://docs.langchain.com/oss/python/integrations/chat/openai): ChatOpenAI(base_url=…) — the legacy openai_api_base is still accepted. (verified 2026-08-06)
- [Vercel AI SDK](https://ai-sdk.dev/providers/openai-compatible-providers): createOpenAICompatible({ name, baseURL, apiKey }) from @ai-sdk/openai-compatible. (verified 2026-08-06)
- [Open WebUI](https://docs.openwebui.com/getting-started/quick-start/connect-a-provider/starting-with-openai-compatible): Admin Settings → Connections → OpenAI → Add Connection, fields URL and API Key (or OPENAI_API_BASE_URL). (verified 2026-08-06)
- [LibreChat](https://www.librechat.ai/docs/quick_start/custom_endpoints): An endpoints.custom entry in librechat.yaml with baseURL, apiKey and models. (verified 2026-08-06)

### Works with any OpenAI-compatible tool
- n8n (OpenAI credential to base URL + API key)
- Microsoft Foundry (bring-your-own-model connection)
- LangChain (ChatOpenAI base_url)
- Vercel AI SDK (createOpenAICompatible)
- Open WebUI (Admin Settings to Connections to OpenAI)
- LibreChat (endpoints.custom in librechat.yaml)
- Dify (OpenAI-compatible provider)
- CrewAI, LlamaIndex, PydanticAI, Mastra, TanStack AI
- Claude Code, Codex CLI, Cline, Roo Code, OpenCode, Continue (agent CLIs)
- Make.com (HTTP module, 40s default timeout noted)

## Comparisons
- [Frontière AI vs openrouter](https://getfrontiereai.eu/vs/openrouter): jurisdiction, pricing, data retention — each row sourced from the competitor's primary documentation
- [Frontière AI vs together-ai](https://getfrontiereai.eu/vs/together-ai): jurisdiction, pricing, data retention — each row sourced from the competitor's primary documentation
- [Frontière AI vs groq](https://getfrontiereai.eu/vs/groq): jurisdiction, pricing, data retention — each row sourced from the competitor's primary documentation
- [Frontière AI vs fireworks-ai](https://getfrontiereai.eu/vs/fireworks-ai): jurisdiction, pricing, data retention — each row sourced from the competitor's primary documentation
- [Frontière AI vs venice-ai](https://getfrontiereai.eu/vs/venice-ai): jurisdiction, pricing, data retention — each row sourced from the competitor's primary documentation

## Blog
- [DeepSeek V4.1 Flash: benchmarks, price, API](https://getfrontiereai.eu/blog/deepseek-v4-1-flash) (2026-09-11): DeepSeek V4.1 Flash, released 2026-09-10: 552B multimodal MoE, 1M context, MIT. What the technical report claims, what we measured on the gateway, and what it costs.
- [Best OpenRouter alternatives in 2026](https://getfrontiereai.eu/blog/best-openrouter-alternatives) (2026-09-02): Honest comparison of OpenRouter alternatives: Together AI, Groq, Fireworks, Venice AI and an EU-sovereign option. Price, jurisdiction and data handling.
- [Best AI API Aggregator for Companies](https://getfrontiereai.eu/blog/best-ai-api-aggregator-for-companies) (2026-08-27): Choosing an AI API aggregator for production: data sovereignty, compliance, billing and model quality. A buying guide for European teams.
- [Top Open-Source and Self-Hosted AI Gateways](https://getfrontiereai.eu/blog/top-open-source-self-hosted-gateways) (2026-08-27): LiteLLM, Ollama, vLLM, TGI, xinference: the best open-source LLM gateways compared — and when a managed gateway like Frontière AI makes more sense.
- [What is Open-Weight AI? A clear guide](https://getfrontiereai.eu/blog/what-is-open-weight-ai) (2026-08-27): Open-weight models: what they are, how they differ from open-source and closed AI, and why enterprises are making the switch.
- [What is Open Source AI? The enterprise guide](https://getfrontiereai.eu/blog/what-is-open-source-ai) (2026-08-27): Open source AI explained: why enterprises are switching from closed models, the best open-source models in 2026, and how to get started.
- [French open-source AI: actors and sovereignty](https://getfrontiereai.eu/blog/ia-open-source-francais) (2026-08-27): French open-source AI models and players: Mistral AI, Hugging Face, Frontière AI, Fleece AI — why Europe is building its own AI ecosystem independent of US tech giants.
- [What is an AI API aggregator? A 2026 comparison](https://getfrontiereai.eu/blog/what-is-an-ai-api-aggregator) (2026-08-26): Frontière AI, OpenRouter, Together AI, Groq, Fireworks AI: which AI API aggregator fits your needs — and why none of the others are GDPR-compliant.
- [Is OpenAI GDPR-compliant? What EU firms need to know](https://getfrontiereai.eu/blog/is-openai-gdpr-compliant) (2026-08-14): OpenAI can be GDPR-compliant in practice, but it is a US company subject to the CLOUD Act — and that is a jurisdictional fact, not a setup choice.
- [Best open-source reasoning models in 2026](https://getfrontiereai.eu/blog/best-open-source-reasoning-models-2026) (2026-08-13): Open-weight models that think before answering: what reasoning actually costs, which models exist, and how to pick one for your workload.
- [Open-source models with tool calling (measured)](https://getfrontiereai.eu/blog/open-source-models-with-tool-calling) (2026-08-12): Tool calling turns a model into an agent. Here is why it is a property of the serving stack, not the weights — and how to verify it before you build.
- [Best open-source embedding models for RAG in 2026](https://getfrontiereai.eu/blog/best-open-source-embedding-models-for-rag) (2026-08-11): Dimensions, context, cost: the measured facts that decide which open-source embedding model to index your RAG corpus with — and why to lock it in early.
- [Self-hosting an LLM vs. calling an API: the numbers](https://getfrontiereai.eu/blog/self-hosting-vs-api-cost) (2026-08-06): A rented H100 costs the same whether it serves one token or a billion. Here is the break-even against per-token pricing, and how to compute your own.
- [Best open-source AI models for EU businesses, 2026](https://getfrontiereai.eu/blog/best-open-source-ai-models-2026) (2026-08-03): GLM-5.3, Qwen3.8, DeepSeek V4 Flash, Kimi K3, Llama 3.3, Gemma 3: which open-weight models are worth deploying, what they cost, and which run on EU sovereign infrastructure.
- [Cloud and GDPR: why a US EU data centre isn't enough](https://getfrontiereai.eu/blog/gdpr-vs-cloud-act) (2026-08-03): A US company's EU data centre does not put your prompts outside US jurisdiction. The legal mechanism, and how to verify a provider's real exposure.

RSS: https://getfrontiereai.eu/blog/feed.xml

## Developer
- [API documentation](https://getfrontiereai.eu/docs)
- [OpenAPI spec](https://getfrontiereai.eu/openapi.json): machine-readable API specification (OpenAPI 3.1)
- [MCP server /api/mcp](https://getfrontiereai.eu/docs): Model Context Protocol for AI agents — list models, get details, search (stateless HTTP, JSON-RPC 2.0)
- [Agent instructions](https://getfrontiereai.eu/agent-instructions): when to use Frontière AI and how to call it (text/markdown)
- [API reference (models)](https://getfrontiereai.eu/api/v1/models)
- [Web console](https://getfrontiereai.eu/dashboard/chat): test models in your browser
- [Public preview](https://getfrontiereai.eu/preview): explore without an account
- [API keys](https://getfrontiereai.eu/dashboard/keys): create and manage keys
- [Billing](https://getfrontiereai.eu/dashboard/billing): balance, usage, top up
- [Sitemap](https://getfrontiereai.eu/sitemap.xml)
- [RSS feed](https://getfrontiereai.eu/blog/feed.xml)

## Company
- [Home](https://getfrontiereai.eu)
- [About](https://getfrontiereai.eu/about): who we are, our mission, FLEECE AI SASU
- [Contact](https://getfrontiereai.eu/contact): email, response times, support
- [Privacy policy](https://getfrontiereai.eu/privacy): GDPR-compliant data processing, zero prompt retention
- [AI Act readiness](https://getfrontiereai.eu/ai-act)
- [Pricing](https://getfrontiereai.eu/pricing)
- [Blog](https://getfrontiereai.eu/blog)