Apertus is an open foundation model developed by EPFL and ETH Zurich as a sovereign AI alternative to closed models like Claude and GPT.
Released in September 2025 and updated to v1.1 in June 2026, Apertus publishes all training data, weights, code, and alignment methods publicly.
What Is the Apertus Open Foundation Model?
Apertus is developed by the Swiss AI Initiative, a collaboration between EPFL, ETH Zurich, and CSCS (the Swiss National Supercomputing Centre).
The model competes with 8B and 70B parameter open-source models and is trained natively on over 1,800 languages with 40% non-English content.
Unlike most open-source models that release weights without full reproducibility, Apertus publishes its complete training data and methods.
The Apertus team designed the model with Swiss data protection law and EU AI Act compliance as foundational principles from the start.
The latest release, Apertus v1.1-4B-Instruct, is optimized for instruction-following tasks and was released in June 2026.
Why Apertus Sovereign AI Became Urgent in June 2026
Sovereign AI moved from policy theory to urgent business need when the US Department of Commerce restricted Anthropic’s Fable 5 and Mythos 5 globally.
Unable to verify user nationality in shared cloud infrastructure, Anthropic disabled both models worldwide, cutting off international users instantly.
This single event demonstrated that organizations relying entirely on US-controlled AI models have zero sovereignty over their own AI access.
As we reported in our piece on the Anthropic export ban, the outage affected companies and governments on six continents without warning.
Apertus, running on Swiss infrastructure with fully open weights, cannot be switched off by any single government’s export control order.
Apertus vs Claude and GPT: Key Differences for Enterprise Deployment
Apertus is fully open and self-hostable, meaning enterprises can run it on their own servers with no dependency on Anthropic or OpenAI.
Claude and GPT are closed models accessed via API, meaning any access restriction, pricing change, or outage is entirely outside your control.
Apertus scores lower than Claude Sonnet 4.6 on coding and reasoning benchmarks, but the gap is narrowing with each model release.
For regulated industries like healthcare, finance, and government, the ability to audit training data is a compliance advantage Apertus alone offers.
Per AI News coverage, Switzerland’s neutrality also makes Apertus politically attractive to non-Western governments seeking AI independence.
Apertus at ACL 2026: What the Research Paper Reveals
The Apertus research paper will be presented at the ACL 2026 Main Conference in San Diego from July 2 to 7, 2026.
The paper details the training methodology, multilingual data curation process, and the alignment techniques used to build a compliant AI system.
Key findings include that Apertus’s multilingual training significantly outperforms monolingual models on cross-lingual transfer and low-resource languages.
See our big tech AI investment analysis for context on how the AI sovereignty movement is reshaping enterprise strategy.
The open publication of training methods means any organization can validate Apertus’s safety and alignment claims independently, unlike closed models.
Related Articles
Trust Post Desk
A journalist and editor at TrustPost.org covering world and national news, technology updates and human-interest stories. They check every fact, interview sources in person or online, and aim to deliver clear, accurate reporting. Their work ranges from breaking news to in-depth features and daily newsletters. Outside the newsroom, they follow emerging trends and engage with readers on social media.
