gpt-oss Hosting in Germany: OpenAI's Open-Weight Models on a Dedicated GPU
You want to use a reasoning model from OpenAI without sending data to the OpenAI API. We run gpt-oss-120b or gpt-oss-20b on a dedicated GPU server in a German data centre.
Companies worldwide trust WZ-IT
Why run gpt-oss yourself
gpt-oss are OpenAI's open-weight models. Both variants are mixture-of-experts models with adjustable reasoning effort and tool calling. The expert weights were trained in MXFP4 format, so the checkpoints are small for their parameter count.
Running gpt-oss yourself means using the model without any connection to OpenAI. Costs depend on the server, not on consumed tokens, and data does not leave the German data centre through a model API.
Data stays in Germany
The model runs on a dedicated server in a German data centre. Under a data processing agreement, with no data path to a model API.
Predictable costs
A fixed monthly price per server tier instead of billing per consumed token. More requests do not increase the invoice.
Model version under control
Checkpoint and quantisation stay until you agree to a change. No silent model update by a provider.
gpt-oss checkpoints at a glance
Vendor specifications and weight file size per format, with source.
| Checkpoint | Released | Parameters | Context (vendor) | Weights per format |
|---|---|---|---|---|
| openai/gpt-oss-120b | 08/2025 | 117B total / 5.1B active (MoE) | 131,072 tokens |
|
| openai/gpt-oss-20b | 08/2025 | 21B total / 3.6B active (MoE) | 131,072 tokens |
|
GB = 10⁹ bytes, sum of the weight files in the listed Hugging Face repository. Operation additionally needs memory for KV cache, runtime and image processing where applicable. As of October 2026.
gpt-oss: licence and commercial use
- Licence
- Apache License 2.0 + gpt-oss Usage Policy (Licence text, Usage policy)
- Commercial use
- Yes, without a revenue or user threshold.
- Obligations and thresholds
- Include the licence text and notices when redistributing. The bundled usage policy requires compliance with applicable law.
- Checked
- As of October 2026
Technical classification, not legal advice. The licence text of the deployed model version is authoritative.
Open LLM licences comparedMinimum recommended server tier for gpt-oss
The recommendation follows from the weight size plus headroom for context and runtime.
| Model | Format | Weights | Minimum recommended | Note |
|---|---|---|---|---|
| gpt-oss-120b | MXFP4 | 65.3 GB | Managed GPU Server 96on premises: AI Cube (128 GB unified memory) | For on-premises operation, the AI Cube with 128 GB unified memory is the alternative. |
| gpt-oss-20b | MXFP4 | 13.8 GB | Managed GPU Server 24 |
Which tier suits your use case depends on context length and concurrent requests. We check this before the proposal.
The Managed GPU Server tiers
Relevant for this family
Managed GPU Server 24
1 × RTX PRO 4000 Blackwell, 24 GB GPU memory
€699 excl. VAT / month, cancellable monthly
€499 excl. VAT one-time setup
View configurationRelevant for this family
Managed GPU Server 96
1 × RTX PRO 6000 Blackwell Max-Q, 96 GB GPU memory
€1,799 excl. VAT / month, cancellable monthly
€999 excl. VAT one-time setup
View configurationManaged GPU Server 192
2 × RTX PRO 6000 Blackwell Max-Q, 192 GB GPU memory
Price and term on request
View configurationManaged GPU Server 384
4 × RTX PRO 6000 Blackwell Max-Q, 384 GB GPU memory
Price and term on request
View configurationThe Managed GPU Server 288 with three GPUs is intended for several models side by side, because vLLM only splits a model when the attention heads are divisible by the number of GPUs.
Running gpt-oss with vLLM
Architecture, quantisation and distribution across several GPUs.
Architecture
Both models use GptOssForCausalLM, an architecture supported by vLLM. gpt-oss-120b has 128 experts and gpt-oss-20b has 32, with 4 active per token in each.
MXFP4
The expert weights are stored in MXFP4, attention and embeddings in BF16. We determine which kernel runs on the RTX PRO 6000 Blackwell with the deployed vLLM version and verify it before the proposal.
Tensor parallelism
With 64 attention heads and 8 KV heads, gpt-oss can be split across two or four GPUs. A single 96 GB GPU is usually sufficient for gpt-oss-120b; more GPUs add room for longer contexts and more parallel requests.
Reasoning and format
gpt-oss uses the harmony response format with a separate reasoning channel. vLLM exposes it through the OpenAI-compatible API. You choose the reasoning effort (low, medium, high) per request.
What operation includes
The Managed GPU Server is an operated model environment, not an empty server.
Model and vLLM
An agreed model in the agreed quantisation, served through a managed vLLM inference layer.
OpenAI-compatible API
Applications and coding clients connect to the server via base URL and API key.
Open WebUI
A managed chat interface for teams using the model without their own application.
24/7 monitoring
Host, GPU, vLLM and Open WebUI are monitored proactively; incidents are handled according to the service level.
Updates with CVE assessment
Operating system, drivers, vLLM and Open WebUI are reviewed and updated in a controlled way. Model changes only by agreement.
DPA and documentation
Data processing agreement, documented configuration and a personal point of contact.
Scope, prices and multi-GPU servers are on the product page.
View Managed GPU ServerCompare other model families
Qwen hosting
Qwen
Qwen hosting in Germany: Qwen3.8-27B, Qwen3.5-122B and Qwen3-Coder-Next on a dedicated GPU with vLLM, an OpenAI-compatible API and operation under a DPA.
View familyGemma hosting
Gemma
Gemma hosting in Germany: Google's Gemma 4 31B, 26B A4B and 12B on a dedicated GPU with vLLM, an OpenAI-compatible API and operation under a DPA.
View familyMistral hosting
Mistral
Mistral hosting in Germany: Mistral Small 4, Devstral Small 2 and Mistral Medium 3.5 on a dedicated GPU, with vLLM, licence review and operation under a DPA.
View familyDevstral hosting
Devstral
Devstral hosting in Germany: Mistral AI's Devstral Small 2 and Devstral 2 for coding agents, with vLLM, an OpenAI-compatible API and operation under a DPA.
View familyDeepSeek hosting
DeepSeek
DeepSeek hosting in Germany: DeepSeek-V4-Flash on a dedicated four-GPU server with vLLM, under the MIT licence and with no data path to the DeepSeek API.
View familyLlama hosting
Llama
Llama hosting in Germany: why Llama 4 is ruled out by its licence for companies in the EU, how Llama 3.3 70B is operated and which alternatives exist.
View familyModel assessment
Have your model and server tier assessed
Name the model, context length and concurrent requests. We check checkpoint, licence and server tier before the proposal.
Frequently asked questions about gpt-oss hosting
Licence, server tier, data path and operation
gpt-oss-120b is the larger model for demanding reasoning and agent tasks; the minimum recommended tier is the Managed GPU Server 96. gpt-oss-20b is intended for simpler tasks, classification and fast answers; the minimum recommended tier is the Managed GPU Server 24.
No. gpt-oss are separate open-weight models that OpenAI published for self-hosting. With hosting there is no connection to OpenAI. Features and answer quality do not automatically match the models in ChatGPT.
Yes. The models are licensed under Apache 2.0, supplemented by a short usage policy that requires compliance with applicable law. There is no revenue or user threshold.
In many cases, yes. vLLM provides an OpenAI-compatible API; applications switch via base URL, API key and model name. Features only offered by the OpenAI platform, such as hosted tools, are checked case by case.
Yes, the AI Cube with 128 GB unified memory is an alternative to the hosted server. Whether it covers your use case depends on context length and concurrent requests. We check this before the proposal.
gpt-oss supports up to 131,072 tokens. The configured context length depends on the server tier and parallelism and is stated in the proposal.
The entry point for gpt-oss is the Managed GPU Server 24, the minimum recommended tier for gpt-oss-20b: €699 excl. VAT / month plus €499 excl. VAT one-time setup, including the dedicated server, vLLM, Open WebUI and managed operation by WZ-IT. For on-premises operation inside your own network, there is the AI Cube. Which tier fits your use depends on context length and concurrent requests. We check this before the proposal.
Let's Talk About Your Idea
Whether a specific IT challenge or just an idea - we look forward to the exchange. In a brief conversation, we'll evaluate together if and how your project fits with WZ-IT.
Arrange a callback
Companies worldwide trust WZ-IT
