Apertus 1.5, the latest iteration of the open-source model series developed by Swiss AI, extends both the 8B and 70B parameter models with significant new capabilities. Released under an Apache 2.0 license to Hugging Face on July 24, 2026, the update introduces native image and audio understanding, an optional thinking mode for complex reasoning tasks, and a context window that has been quadrupled to 262,144 tokens. The models were trained through continued pretraining, with the 8B variant receiving 4 trillion tokens of text and multimodal data and the 70B receiving 2 trillion tokens, while maintaining over half of training data as text-only to preserve core language abilities. The technical improvements span both architecture and training methodology. Images are tokenized using the Emu3.5 tokenizer and audio through WavTokenizer, integrating multimodal inputs directly into the text sequence rather than using separate encoders. Post-training incorporates five stages including supervised fine-tuning, reinforcement learning with verifiable rewards, and alignment with the Apertus Charter. The release includes comprehensive evaluation results showing competitive performance in multilingual tasks, legal document understanding, and machine translation, with the 8B model matching GPT-OSS-20B on translation benchmarks and outperforming Claude Haiku 4.5. Apertus 1.5 maintains the project's commitment to transparency by releasing open weights, open training data, and full technical documentation. All training sources are publicly available and permissively licensed, with robots.txt opt-outs respected retroactively, reinforcing the project's position in the growing ecosystem of open-source alternatives to proprietary large language models.