Apertus Releases 1.5 with Multimodal Capabilities and 262K Context Window
Key Takeaways
- ▸Apertus 1.5 adds multimodal image understanding, thinking mode, and expands context window to 262,144 tokens—4x larger than Apertus 1.0
- ▸Both 8B and 70B models released with fully open weights, training data, and complete training methodologies—maintaining Apertus's full transparency commitment
- ▸Models demonstrate improved instruction-following and tool use, with support for text, images, and experimental audio inputs
Summary
Apertus has released version 1.5 of its open-source language models, available in 8B and 70B parameter sizes as of July 24, 2026. The update extends the previous Apertus 1.0 models with significant new capabilities including native image understanding, an optional thinking mode for improved reasoning, a four-fold increase in context window to 262,144 tokens, and enhanced instruction-following and tool use. The 8B model was trained on 4 trillion additional tokens while the 70B received 2 trillion tokens of multimodal training data.
Staying true to its open philosophy, Apertus 1.5 releases with fully open weights, open training data, and complete transparency around training details and values. The models support text, images, and experimental audio input, enabling tasks like technical document analysis and diagram interpretation. The company is coordinating with inference providers to make the models available across multiple platforms.
Apertus emphasizes improved instruction adherence and better integration with external tools and APIs compared to the 1.0 release. A technical report with detailed benchmarks, training pipelines, and intermediate checkpoints will be published in the coming weeks, with model cards and setup instructions already available on Hugging Face.
- Availability expanding across inference platforms with technical documentation and community engagement through Hugging Face forums
Editorial Opinion
Apertus 1.5 represents a meaningful commitment to truly open AI development. By releasing not just model weights but complete training data and methodologies, Apertus demonstrates that open-source AI can scale to competitive capability levels without sacrificing transparency. The addition of multimodal capabilities brings Apertus closer to feature parity with leading proprietary models while maintaining its core values of openness and accountability.
