IONOS AI Model Hub Prices: How Much Does AI from German Data Centers Cost?

The IONOS AI Model Hub bills per million tokens processed, separated by input and output, with no basic fee. Prices range from €0.10 (input, e.g. Qwen3.5-9B or Mistral Small 24B) to €3.60 (output, Qwen3.5-397B) per million tokens; Embedding models for document search cost €0.01 to €0.10 (net, according to the IONOS price list, as of October 2026).

If you want to use AI in your company without giving data to US providers, you will quickly end up with IONOS. The price list is clear – the more difficult question is what this means in practice. That's what this is all about.

What is a token?

A token is a text module that a language model processes – roughly a syllable or a short word. The tokens that you send (input: question, instructions, document passages sent along) and the tokens that the model writes (output) are billed.

The prices at a glance (net, per 1 million tokens)

In addition, there are special models:

• Embeddings (for document search): €0.01 (paraphrase-multilingual-mpnet-base-v2) to €0.10 (Qwen3-VL-Embedding-8B) per million tokens
• Reranker Qwen3-VL-Reranker-8B: €0.04 per million tokens
• Text recognition LightOnOCR 2: €0.15 input, €0.30 output
• Image models: e.g. FLUX.1 [fast] €0.0288 per image

The range of models and prices change regularly. For binding calculations, the current price list of IONOS applies.

Calculation example: Chatbot on the website

Example: A website chatbot answers 1,500 questions per month. With instructions and text passages found, around 3,000 tokens go in and 300 tokens go out per question. That's 4.5 million input and 0.45 million output tokens. With Mistral Small 24B, this costs around €0.45 plus €0.14, i.e. less than €1 net per month; with Llama 3.3 70B, around €3.20. In addition, there are the one-time embeddings for the knowledge base – an amount in the cent range for 2,000 pages.

The model costs are therefore rarely the bottleneck. More expensive are the construction, maintenance of the content and operation of the application around it.

What drives up costs

• Long entries: If you send entire documents instead of suitable excerpts, you pay many times more.
• Large models for simple tasks: A 397B model for email sorting is a waste.
• Long answers: Issuance tokens are often more expensive than input tokens.
• Automated mass processing: Thousands of documents per day add up.

Comparison with your own hardware

In-house hardware is only worthwhile when the volume is high, even or when data is not allowed to leave the house at all. For most SMEs, the way via the AI Model Hub is cheaper and more flexible. The article Hardware for your own AI helps with the consideration.

We integrate the IONOS AI Model Hub into chatbots, document searches, and workflows and estimate in advance what costs you can expect.

Counselling and accompanied implementation is eligible for funding: Through INQA coaching, the federal government covers 80% of the consulting costs of up to €1,200 per day for up to 12 days, i.e. up to €11,520 subsidy. In Mecklenburg-Western Pomerania and the other new federal states, BAFA consulting funding is also eligible at 80% until 31.12.2026 (50% in the old federal states). Funding is provided for consulting, not software. The funding check shows which program fits.

Frequently Asked Questions

How much does the IONOS AI Model Hub cost?

Billing is per million tokens, depending on the model from €0.10 to €3.60 (net, as of October 2026). There is no basic fee.

Is there a basic fee for the IONOS AI Model Hub?

No. You only pay for the tokens or images that are actually processed.

Which model makes sense to get started?

For many tasks, Mistral Small 24B is a good starting point: inexpensive and versatile. For more difficult tasks, gpt-oss-120b or Llama 3.3 70B come into question.

Are the prices net?

Yes, the price list lists net prices plus VAT.

Would you like to estimate the costs for your use case? Get in touch via the contact page.