Apertus 1.5 (apertus-ai.org)

Apertus 1.5 has been launched, advancing the capabilities of its predecessors, the 8B and 70B models, with key enhancements that are poised to transform interactions in the AI/ML landscape. This update introduces native image understanding, allowing the models to analyze and respond to images, documents, and even spoken language, albeit in an experimental phase. The "thinking mode" feature enables the models to reason through inputs before generating answers, which significantly boosts their performance on complex reasoning tasks. Furthermore, the context window has expanded to an impressive 262,144 tokens, quadrupling the previous limit and enabling richer and more context-aware interactions. The model's architecture is further refined with improvements in instruction adherence and tool integration, enabling more accurate responses and seamless collaboration with external applications. With the addition of 4 trillion and 2 trillion tokens of text and multimodal data for the 8B and 70B models, respectively, the training process emphasizes transparency with open weights, data, and values. These enhancements ensure that Apertus 1.5 not only offers advanced technical capabilities but also aligns with the principles of community involvement and ethical development. Detailed benchmarks and technical instructions will be shared soon, inviting developers and researchers to engage with the platform actively.