Model catalogue · BAAI
BGE Embedding
The most widely used open embedding model for multilingual search: it delivers dense, keyword-style and passage-level representations from a single model and stays economical enough to run alongside the language model on the same graphics card. For professional secrets the chain is closed on two routes - via the Open Telekom Cloud, because the model is operated there by the operator itself in German data centres and the signed agreement under section 203 of the German Criminal Code covers precisely these services, and in dedicated operation on reserved hardware, where only the data centre operator participates and is bound via section 203(4). Via IONOS the product-specific confirmation of the confidentiality clause is so far missing.
Everything section 203 requires is settled, including the provider's confidentiality undertaking. You do not have to clarify anything further.
What it is good for
- Available on three German routes - via Telekom, via IONOS and on the dedicated instance
- 0.6 billion parameters in BF16, around 4 GB - runs alongside the language model on the same card
- Three representation types from one model, multilingual, under Apache 2.0
Where it stops
- Embedding is a side role - it arranges the corpus, it answers no question
- The entire file corpus runs through this model, not just the current request - the contractual check weighs more heavily here than for an occasionally used language model
- No legal fine-tuning; judging the relevance of hits stays with the practitioner
Routes examined
The same model, procured differently, is a different legal position. What is examined per route is the contract chain, not the model.
| Route | Section 203 status | Version on this route |
|---|---|---|
| Dedicated instance in Germany, operated by Gosign | covered | - |
| IONOS AI Model Hub (German data centres) | conditional | - |
| German data centres through the Gosign chain | covered | - |
The conditions behind a conditional status - which piece is open on which route and who closes it - are set out in the full matrix in the German catalogue, and we go through them in the conversation. The German page for BGE Embedding carries the sources with their retrieval dates.
Is BGE Embedding the right fit?
We go through your tasks and the route that carries them - and say plainly where it does not.
Arrange a conversation