The Swiss AI Initiative has published release notes for Apertus 1.5 on Hugging Face. The update covers the 8B and 70B models from the Apertus 1.0 generation. The weights went up on July 24, 2026 under an Apache 2.0 license, and the notes, dated October 15, 2026, describe the technical report behind them.
What changed
- Image and audio input. The models accept images and audio alongside text, so documents, diagrams and photos can go into the same conversation. Spoken-language processing is labelled experimental.
- Thinking mode. An optional mode lets the model reason about the input before answering. The authors say this improves results on reasoning problems.
- Longer context. The window is now 262,144 tokens, four times the Apertus 1.0 size.
- Instruction following and tool use. The team reports more predictable adherence to instructions. Tool-use training data was limited to trajectories verified by execution.
The models understand images and audio but generate only text. Images and audio enter as tokens in the same sequence as text. A 512×512 image becomes 1,024 tokens, and audio costs 40 tokens per second. The architecture is unchanged and only the vocabulary was extended.
How it compares
The authors say plainly that Apertus 1.5 still trails frontier open-weight models. They report clear gains in coding and instruction following. The 8B model outperforms similarly sized open-weight models on instruction-following and multilingual tasks. On multimodal work it sits closer to Gemma 4 12B and other current open-weight models. It is competitive on spatial understanding among open-weight models, and on document and chart understanding among open-source models.
Two external results stand out. On Swiss legal tasks, the model is on par with leading open-weight models. On the Last Translation Benchmark, the 8B model matches GPT-OSS-20B and beats Claude Haiku 4.5 and Google Translate. These are the authors' reported figures and were not independently checked here.
Where to get it
The weights and intermediate checkpoints (base, SFT, RLVR and DPO) are on Hugging Face, along with training data mixes. Proton Lumo and Infomaniak Euria have integrated Apertus 1.5. Swisscom, Infomaniak, Phoeniqs and other providers have announced API access. The project also lists the evaluation results in a shared format intended to make them reproducible and comparable.
What comes next
The team says it announced a roadmap for Apertus 2.0 at the AI+X Summit in Zurich. It describes a sparse or mixture-of-experts architecture of about 600B total parameters, with agents, mathematics and coding as priorities. That is a plan, not a release.