Swiss AI published Apertus 1.5 release details on October 15. The 8B and 70B models can accept images and audio as discrete tokens in the same sequence as text, understanding documents, charts, photos and experimental speech while generating text only. The context window grows to 262,144 tokens, with optional thinking mode and improved instruction following and tool use. Continued training added 4 trillion tokens to the 8B model and 2 trillion to the 70B model. Weights, data sources, recipes, intermediate checkpoints and reproducible evaluations are public. The team says overall performance still trails frontier open-weight models; the distinctive value is transparency, auditability and local deployment rather than a claim of benchmark dominance.