Free AI Inference for Hetzner
Hetzner is experimenting with AI inference service.
Right now, you can go ahead and use their AI inference for free.
How to get started?
They’re providing this OpenAI chat completion format.
Base URL: https://inference.hetzner.com/api/v1
Since it’s an OpenAI API format, you can just replace the base url and start using this instantly.
from openai import OpenAI
client = OpenAI(
base_url="https://inference.hetzner.com/api/v1",
api_key="<YOUR_TOKEN>",
)
And you can start working like this
response = client.chat.completions.create(
model="DeepSeek-V4-Flash-0731",
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Why sky is blue?"},
]
)
print(response.choices[0].message.content)
And the following models are available
| Model | Type | Context Length | Modalities |
|---|---|---|---|
| DeepSeek-V4-Flash-0731 | MoE, 304B total / 13B active | 512,000 tokens | Text |
| GLM-5.2-NVFP4 | MoE, 744B total / 40B active | 512,000 tokens | Text |
| Kimi-K2.7-Code | MoE, 1T total / 32B active | 262,144 tokens | Text, Image |
| Qwen/Qwen3.6-35B-A3B-FP8 | MoE, 35B total / 3B active | 262,144 tokens | Text, Image |
Notes
- They’ve 10M input tokens and 200k output tokens per 60 seconds per API key rate limiting
- Inference API is currently in experimental, so do not use it for anything serious
- I think, you might need to have an active account with them to use this API inference. If you’re using their services like VPS then great, if not you might need to do ID verifications and things.