Open models on a GPU of your own,
served from Poland.
The model you pick, on a dedicated GPU, behind an OpenAI-compatible /v1 endpoint. One tenant per GPU — your prompts never share memory with anyone else’s. Hosted in Poland, inside the EU.
- dedicated GPU
- OpenAI-compatible /v1
- vLLM
- Poland · EU
- billed per hour
Models we serve newest first
| Model | Developer | Parameters | Context | Licence | Serving | |
|---|---|---|---|---|---|---|
| Qwen 4.0 27B | Alibaba Qwen | not published yet | not published yet | not published yet | not released yet · register interest | Details → |
| Bielik 11B v3.1 | SpeakLeash · ACK Cyfronet AGH | 11.2B | 32,768 | Apache 2.0 | BF16 · FP8 · NVFP4 · DFlash draft | Details → |
| Qwen3.8 Flash Next | Alibaba Qwen | 180.0B | 262,144 | Qwen Community 1.0 | FP8 · 2× DGX Spark · needs Qwen’s MaaS licence | Details → |
| Qwen3.8 27B | Alibaba Qwen | 27.8B | 262,144 | Apache 2.0 | NVFP4 · MTP · one RTX 5090 | Details → |
How it runs same for every model
- Endpoint
- OpenAI-compatible /v1 — chat, completions, streaming.
- Tenancy
- One customer per GPU. No shared batches, no other tenant in memory.
- Location
- Poland, EU — on machines we own and run.
- Billing
- Per hour of GPU time, USD + VAT. Longer blocks quoted on request.
- New models
- Any open-weight model that fits the memory — ask and it is added here.
A model that is not on the list?
If the weights are open and fit one of the machines, it can run here. Write which model and what load you expect — reply within one business day.