We introduce Apertus, a fully open suite of large language models (LLMs) built to solve two major gaps in open model development: data compliance and multilingual inclusion. Trained exclusively on openly available data—while respecting content rights and filtering for non-permissive or sensitive content—Apertus supports over 1,800 languages, with ~40% of tokens from non-English sources. Released at 8B and 70B scales, the models deliver near state-of-the-art performance on multilingual benchmarks. All artifacts, including data pipelines, training code, and evaluation tools, are openly licensed for full transparency and reuse.