Apertus released open-source language models (8B and 70B versions) on July 24, 2026, with new capabilities including image understanding, a thinking mode for reasoning, and a context window extended to 262,144 tokens—four times longer than the previous version. All model weights, training data, and methodology are publicly available, allowing developers to use and modify the software freely.
Summaries like this, in your inbox every morning.
Sign up free →What happened
Apertus released version 1.5 of its open-source language models on July 24, 2026, with 8B and 70B variants. The update adds image understanding, an optional thinking mode for reasoning, a context window extended to 262,144 tokens (four times longer than Apertus 1.0), and improved instruction-following and tool use. The 8B model received 4 trillion tokens of additional training data and the 70B model received 2 trillion tokens.
Why it matters
Apertus publishes its models as fully open — weights, training data, and methodology included — meaning developers can use and modify the software without commercial licensing restrictions. The multimodal and reasoning enhancements make the models more practical for real-world tasks like document analysis and complex problem-solving, while the longer context window allows the models to work with much larger inputs.
What to watch
A technical report with detailed benchmark results, training pipelines, and intermediate checkpoints will be published in the coming weeks. The models are available on Hugging Face, and the team is working with inference providers to make them available on multiple platforms.
Apertus, a fully open-source AI model project, released version 1.5 on July 24, 2026, delivering two model sizes—8B and 70B parameters—with a suite of new capabilities. The 8B model underwent training on an additional 4 trillion tokens of text and multimodal data, while the 70B variant received 2 trillion tokens, building on the foundation of Apertus 1.0.
The release introduces five key enhancements. First, native image understanding allows the models to accept images alongside text, enabling conversation about documents, diagrams, and photos; experimental support for spoken language processing is also included. Second, a thinking mode allows the models to reason internally before answering, improving performance on reasoning-heavy problems. Third, the context window—the maximum length of text the model can process in a single input—expands to 262,144 tokens, four times the capacity of Apertus 1.0. Fourth, instruction-following has improved, delivering more predictable and accurate responses. Fifth, tool use capabilities have been strengthened for better integration with external APIs and services. The release also emphasizes stronger adherence to the Apertus Charter, a set of values and principles intended to bring transparency to the model's design.
Accessibility is central to the announcement. The models are fully open-source: weights are published, training data is disclosed, training methodology is detailed, and intermediate checkpoints are provided for inspection. The technical report documenting further details and benchmark results is slated for release within the coming weeks. Instructions for running the models are available in the model cards on Hugging Face. The team is coordinating with inference providers—platforms that host and serve models—to ensure the models become available across multiple services.
The Apertus team is also signaling growth, with multiple engineer positions open at ETH Zurich and EPFL, and is inviting developers to contribute feedback and built projects through a community forum on Hugging Face and GitHub.
Apertus 1.5 represents an incremental but substantive step forward in open-source AI capabilities. The addition of image understanding and reasoning mode—features previously associated with proprietary frontier models—brings practical multimodal and reasoning abilities to fully open models. The four-fold increase in context length (to 262,144 tokens) addresses a concrete constraint for applications requiring longer document processing.
The commitment to open methodology is notable: the announcement promises publication of a technical report with detailed benchmark results, training pipelines, and intermediate checkpoints. This transparency, combined with weights and data being publicly available, sets the release apart from closed-source offerings and enables other teams to build upon and audit the work. The team is actively working to distribute the models across multiple inference platforms, widening accessibility beyond researchers with dedicated hardware.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
