Retention by endpoint
What happens to a prompt on each endpoint. Zero means the body is never written to the database at all.
LLM EU Verified means documented, not legally certified. It is not an AI Act conformity assessment and LLM EU is not an EU body.
| Models | Only zero retention is implemented: prompt and completion bodies are not written to the database. Requests requiring nonzero retention are refused, not silently processed under a different retention setting. | Hosting classes | |
|---|---|---|---|
| Mistral Small 4 (119B-A6B) | zero | fr-par | First-party endpoint served by the mock backend. No real weights run until gpu_enabled is set. |
| Devstral Small 2 (24B) | zero | de-ber | First-party endpoint served by the mock backend. No real weights run until gpu_enabled is set. |
| EuroLLM-9B-Instruct (2512) | zero | de-fra | First-party endpoint served by the mock backend. No real weights run until gpu_enabled is set. |
| EuroLLM-22B-Instruct (2512) | zero | de-ber | First-party endpoint served by the mock backend. No real weights run until gpu_enabled is set. |
| EuroLLM-1.7B-Instruct | zero | de-fra | First-party endpoint served by the mock backend. No real weights run until gpu_enabled is set. |
| Whisper large-v3 | zero | de-fra | First-party endpoint served by the mock backend. No real weights run until gpu_enabled is set. |
Where a field has no source, the card shows it as unverified instead of filling it in. An empty field is information.