Apertus 1.5 Brings Image Understanding and 4x Context Window to Open-Source LLM
Key Takeaways
- ▸Apertus 1.5 released in 8B and 70B sizes on July 24, 2026, with 4x larger context window (262,144 tokens) and native image understanding
- ▸Fully open-source release includes model weights, training data, training code, and performance benchmarks—embodying the Apertus Charter values
- ▸New features include optional thinking mode for reasoning tasks, improved instruction-following, better tool use, and experimental spoken language support
Summary
Apertus has released version 1.5 of its fully open-source large language models, extending both the 8B and 70B variants with significant capability upgrades. The new models introduce native image understanding for processing documents and diagrams, an optional thinking mode for improved reasoning, and a quadrupled context window of 262,144 tokens—four times larger than Apertus 1.0. The update also brings enhanced instruction-following and better tool integration for working with external APIs.
The development represents a major continuation effort: the team added 4 trillion tokens of text and multimodal training data to the 8B model and 2 trillion tokens to the 70B model. True to the Apertus Charter, the release includes full transparency—open weights, open training data, complete training details, and published benchmarks. The models became available on July 24, 2026, with Apertus working with inference providers to expand platform availability. Early demonstrations show the models can interpret technical drawings and process multimodal inputs with strong accuracy.
- Available across multiple platforms through inference providers; technical report and detailed benchmarks coming in the following weeks
Editorial Opinion
Apertus 1.5 represents a meaningful milestone in open-source AI development—delivering capabilities typically associated with closed commercial models while maintaining full transparency on training and values. The commitment to releasing weights, data, and training pipelines sets a valuable precedent for responsible AI development. However, the field moves quickly; Apertus must continue aggressively closing the gap with frontier proprietary models to remain competitive and relevant to developers choosing their infrastructure.
