Berget AI partners with Opper AI to bring EU-hosted serverless inference to one API
Stockholm, Sweden, September 2025. Opper AI and Berget AI are partnering to make EU-hosted, pay-as-you-go inference available directly inside Opper’s Task-Completion API. Teams can now select Berget’s models in the Opper console, keep data inside the EU, and use Opper’s model-agnostic routing, evaluations, and observability to ship reliable AI agents.
Why this matters
EU data residency and sovereignty. Berget AI provides EU-hosted inference for private and public sector organizations. It is built to ensure that data remains on sovereign infrastructure within EU borders and to enable AI applications to also handle sensitive data.
Pick the right model for the task at hand. Opper AI gives one API to run 100+ models, now including Berget’s lineup. Set a model for each task and configure fallbacks to handle rate limits or outages.Â
Compare options in Opper’s Model Benchmarks and Task Benchmarks.
Reliability in production. Opper adds built-in evaluations, routing, and fallbacks so AI agents behave like dependable systems.
What’s available today
At launch, the following Berget-hosted models are live in Opper with EU residency:
- GPT-OSS 120B
- Llama 3.3 70B Instruct
- Mistral Small 3.1 24B Instruct
- Qwen3 32B
“Customers want choice, sovereignty, and reliability. Berget brings EU-hosted inference, and Opper adds task contracts, evals, and switching so teams can deliver with confidence,” said Göran Sandahl, Co-founder and CEO, Opper AI.
“We champion AI sovereignty to unlock innovation in Europe. This partnership puts EU-compliant inference in the hands of Opper’s builders within minutes,” said Andreas Lundmark, Co-founder and CEO, Berget AI.
About Berget AI Berget AI provides EU-hosted inference and compute with an OpenAI-compatible API, pay-as-you-go pricing, and in-house GPU clusters. The company focuses on AI sovereignty and keeping data inside EU borders.
About Opper AI Opper AI is the Task-Completion API that lets developers define tasks once and run them across 100+ models. The platform adds smart prompting, built-in evaluations, routing with fallbacks, and full observability to make model calls as reliable as any other API.