Apertus – Open Foundation Model for Sovereign AI
Points and comments are a snapshot, not live.
Apertus is a fully open foundation model for sovereign AI developed by a Swiss consortium.
Developed by EPFL, ETH Zurich, and CSCS via the Swiss AI Initiative, Apertus offers open weights, data, code, and training recipes. It claims compliance with the EU AI Act, supports 1000+ languages, and is competitive with top open models at 8B and 70B scales. Swisscom is a strategic partner.
What commenters are saying
Commenters are split. Some praise its full openness (open weights, data, training details) as a scientific contribution, others criticize past versions as "pretty bad" and not truly copyright-compliant. Several point to other fully open models like OLMo and K2 Think V2, and note Nemotron's partially proprietary dataset. A user finds Apertus useful for RAG but not agentic use; another reports unreliable multilingual performance.