Embeddings
Turn documents into vectors, in the EU
Frontière AI's embeddings endpoint converts text into vectors for retrieval-augmented generation — search, RAG, semantic routing — over an OpenAI-compatible API. Point your existing OpenAI client at our base URL: your text is embedded by EU-sovereign providers (OVHcloud, Scaleway), through a gateway running in Paris, and every model is labeled sovereign or fast access before you call.
What Frontière AI offers for embeddings
A single POST /api/v1/embeddings endpoint that speaks OpenAI's contract. Send text as a string or an array, get back one vector per input in the order you sent them, authenticated with the same API key as your chat calls and billed from the same prepaid balance.
Endpoint
curl https://getfrontiereai.eu/api/v1/embeddings \
-H "Authorization: Bearer sk-front-…" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen3-embedding-8b",
"input": ["Your document text", "Another sentence"]
}'Built for retrieval-augmented generation
Embeddings are the retrieval half of RAG: you index your documents once, then at query time you embed the question and look up the nearest vectors. Frontière AI fits that pipeline for teams who need the data to stay inside EU jurisdiction:
01
Pick your dimensions before you index — the vector length is part of your storage schema, and changing model later means recomputing every vector.
02
Each embedding model lists its verified context window, so you chunk to fit it instead of guessing.
03
Embeddings are billed on input tokens only — no output tokens, no output rate — so indexing large corpora is cheap and predictable.
Embedding models
Every model below is live. Vector dimensions were measured with a real call; context windows are the limits the provider states — in its API response (OVHcloud) or in its error when we exceeded it (Scaleway). Values we have not verified are simply absent.
| Model | Dimensions | Context | Hosted via | Price /100k in |
|---|---|---|---|---|
| BGE Multilingual Gemma2bge-multilingual-gemma2Open weightsModel card ↗ | 3584 | 8,192 tokens | EU sovereign · Scaleway | €0.0168no output rate |
| Qwen3 Embedding 8Bqwen3-embedding-8bOpen weightsModel card ↗ | 4096 | 40,960 tokens | EU sovereign · Scaleway | €0.0168no output rate |
Why dimensions and context matter
Vector length defines your storage schema and your recall quality — smaller vectors index faster and cheaper, larger ones carry more nuance. The context window sets how long a single text can be before it must be chunked. Both are yours to size up front, because changing them later means re-embedding your whole corpus.
Billing
Embeddings consume input tokens only, at the model's input rate, from your prepaid balance — no subscription, no output rate. The price shown is the client price per 100,000 tokens, margin and VAT included.
FAQ
Can I use the official OpenAI SDK for embeddings?
Yes. Our embeddings endpoint implements OpenAI's contract — point base_url at the Frontière AI API, pass your key as api_key, and client.embeddings.create works unchanged.
Are my embeddings stored anywhere by Frontière AI?
No. Frontière AI only bills the call and logs what billing requires (model, token counts, amount). The vectors themselves live in your system, in your storage, under your jurisdiction — we never hold your index.
Which jurisdiction do embedding models run in?
Each embedding model carries a sovereignty label. Sovereign models run on OVHcloud or Scaleway — providers with no non-EU capital control. No model is ever labeled sovereign unless it meets that standard.
What is the cheapest embedding model?
BGE Multilingual Gemma2 and Qwen3 Embedding 8B both start at the same low input rate (seen on the /pricing page); BGE-M3 is paused since 20 October 2026. The price shown is per 100,000 input tokens, margin and VAT included.
How do I pick dimensions before building my index?
Decide before you index: the vector length becomes part of your storage schema, and changing model later forces a full re-embed. Pick the model whose dimensions and context fit your recall and cost needs, then lock it in.
Index in the EU, retrieve in the EU
Create an account, top up €10, and embed from your existing OpenAI client in minutes.