NEWGLM-5.2, Nemotron 3 and Qwen 3.6 are now in production.
Hosted in Munich and included free with every paid plan for a short time.
The LLM Serving Service gives your teams OpenAI-compatible access to the leading AI models: open-source models hosted by Telekom in Germany, frontier models from OpenAI, Anthropic, Google, and Mistral in EU regions. One contract, one endpoint, one straight answer for your auditor.
Free trial: 5M tokens, 3 months. No credit card required.
It's rarely the models that slow you down. It's everything around them.
Each team builds AI on its own: separate vendors, separate systems, time wasted on multiple compliance checks. Effort put in, but your organization is still miles away from scaling. Nothing is reusable, and every new use case pays the full setup cost again.
Every AI provider triggers the same questions: Is our data safe? What happens to our inputs? What if a security incident occurs? Months pass, competitors ship.
Your teams want GPT today, Claude tomorrow, an open-source model for the sensitive workload. Onboarding each provider separately means a new procurement round every time.
Time to shift gears: from pilot to production.
LLM Serving speaks the OpenAI API standard. If your code works with OpenAI, it works with us: same SDKs, same request format, one configuration change. Behind that one endpoint sits the full catalogue. Switch models by changing one string — including switching between a US frontier model and a Germany-hosted open-source model, per request, per use case.
from openai import OpenAI
client = OpenAI(
api_key=os.environ["AIFS_API_KEY"],
base_url="https://llm-server.llmhub.t-systems.net/v2",
)
response = client.chat.completions.create(
model="gpt-oss-120b",
messages=[{"role": "user", "content": "Hello from sovereign Europe."}],
)All models run in the EU. Restrict your selection to Telekom-hosted models only, or use the full catalogue.
open-source models including GPT-OSS 120B, Qwen3 family, Llama 3.3, Mistral Small, plus German-language embeddings and Whisper speech-to-text. Your data is not stored, not visible to Telekom or third parties, and never used for training.
GPT-5 family on Azure, Claude 4.6 and Gemini 3 on Google Cloud, Mistral Medium — all in the EU. Operated by the respective provider, contracted through Telekom.
Chat, reasoning, coding, vision, speech-to-text, and image generation. New models are added continuously.
Public sector, financial services, and healthcare teams use AIFS in production today.
EU AI Act.
Independently assessed against the EU AI Act.
Sovereign and secure infrastructure.
AIFS runs on T-Cloud public: BSI C5 Type 2 attested and ISO 27001 certified Telekom infrastructure.
GDPR-compliant.
With data processing agreements available and a German contractual counterparty.
Transparent by design.
Server locations, terms of use, and licensing provisions are published per model.
One platform from your first API call to organisation-wide rollout.
The reliability, controls, and support your teams need from day one.
Production-ready service. 99.5% availability, flexible enterprise-grade support.
Self-service portal. Generate and manage API keys, track token usage by key and model, add users — no ticket required.
Scalable by design. Ready to match strict technical requirements, with tenant-separated, dedicated instances where needed.
OpenAI-compatible means the ecosystem just works: the OpenAI SDK, LangChain, LlamaIndex, and coding assistants like OpenCode and Cline connect out of the box. Keep the tooling your teams chose, make it sovereign underneath.
Our team provides professional services around the product: dedicated model instances for exclusive capacity and additional models on request. Fine-tuning and deployment in your own environment are also available as professional services where a use case justifies them.
Common starting points. Yours may be different.
Point OpenCode, Cline, or your internal tooling at Germany-hosted coding models. AI pair programming without sending your source code outside the EU.
Run your voicebots and chatbots on one governed endpoint with per-model compliance choices and usage tracking per API key.
Extraction, classification, and summarization over sensitive documents, with vision-capable models and the option to keep processing on Telekom infrastructure in Germany.
Whisper hosted in Germany: transcribe calls, meetings, and recordings containing personal data without a US data transfer discussion.
Talk to us about your own pain points, application scenarios, and ideas.
Token-based pricing, the same on every plan; higher tiers add Premium models and higher rate limits. Self-service from €1,000/month, or a custom Enterprise agreement. Try it free first, no credit card required.
Your first API call is five minutes away.
