Hetzner Inference (Experiments)
first_party
Getting credentials
api-key
api_key via header (header Authorization: Bearer)
Environment: HETZNER_API_KEY
Create a token in the Experiments dashboard; free while experimental, no SLA.
Endpoints
| ID | URL | Protocol | Auth |
|---|---|---|---|
| chat | https://inference.hetzner.com/api/v1/chat/completions | openai-chat | any |
API surfaces
textstreaming
Quirks
- Experimental 'as is' program (launched ~2026-07): free, EU-based, no production guarantees; email notice before changes. GET /api/v1/models is definitive for the lineup. (docs)
- Limits per key: 10 requests/60s, 4M input and 100k output tokens/60s. Reasoning control is undocumented; a community-reported chat_template_kwargs.enable_thinking flag exists for Qwen3.6. (docs)
Models offered
| Model | Wire ID | Endpoint | Status | Reasoning |
|---|---|---|---|---|
| alibaba/qwen3-8-27b | Qwen/Qwen3.8-27B | chat | preview | toggle · default on · returns reasoning_content |