A new LLM from Europe to challenge American and Chinese models

Switzerland has unveiled Apertus, its first fully open-source multilingual large language model, positioning the Alpine nation as a challenger to American and Chinese AI dominance with a privacy-focused alternative built for transparency and regulatory compliance. The announcement on September 2 marks a milestone in the global AI race, as Switzerland joins the growing movement toward sovereign AI infrastructure.

A LLM from France, Mistral, is already known to some extent. I asked Perplexity to compare them: 

Apertus and Mistral are both open-source large language models, but they diverge significantly in their technical, regulatory, and linguistic approaches. Apertus is distinguished primarily by its comprehensive openness, transparent documentation, and strong privacy/regulatory focus, while Mistral is known for modular deployment options and performance-driven design for enterprise contexts.

Technical Transparency and Documentation

• Apertus offers full transparency: training process, model weights, architecture, data sources, and intermediate checkpoints are all public, enabling full reproducibility and independent audits of model compliance and quality.ggba+1

• Mistral supplies open model weights and documentation, but typically does not fully publish all details about the training corpus, copyright handling, or step-by-step reproducibility; Mistral’s focus is on developer flexibility and modular toolsets, not deep regulatory transparency.walturn+1

Privacy and Regulatory Compliance

• Apertus was “designed to comply with Swiss data protection laws and the EU AI Act,” featuring data curation with removal of personal data and opt-out mechanisms.linkedin+2

  • It is the first LLM to publish compliant summaries of training data and copyright policy per new EU rules.latticeflow+1

  • Specifically filtered for privacy and regulatory obligations before training, reducing risk by never including problematic data in the first place.cyberinsider+1

• Mistral provides open deployment channels (local, cloud, API) for compliance-heavy industries, but its approach to privacy and copyright is less explicit; it does not publish the same level of data or regulatory documentation as Apertus and focuses more on technical utility and low-latency enterprise controls.mistral+1

Multilingual and Cultural Coverage

• Apertus is trained on 15 trillion tokens in over 1,000 languages, with 40% of data dedicated to non-English—including underrepresented languages such as Swiss German and Romansh—demonstrating a unique global and European language emphasis.ggba+1

• Mistral supports multilingual tasks competitively, with strong performance in European and global languages via its general-purpose and vertical-specific models, but its detailed linguistic data composition is not as open; the primary focus is practical use cases rather than linguistic preservation.walturn+1

Enterprise and Open-Source Philosophy

• Apertus is built as sovereign public digital infrastructure: open to commercial and research use, emphasizing accessibility for European and privacy-sensitive organizations.cyberinsider+1

• Mistral is designed for developer and enterprise flexibility, offering modular architectures, vertical-adaptive models (law, medicine, finance), multimodal support (vision, code generation), and fine-tuning via LoRA, RLHF, and supervised data.mistral+1

Summary Table

Apertus stands out for its privacy-first design, regulatory compliance, and openness—particularly for European and public sector needs—whereas Mistral excels in enterprise customization, performance, and developer tooling for broader commercial purposes.latticeflow+3