Apertus 1.5
The Swiss AI Initiative published the Apertus 1.5 technical report. The 8B and 70B open models add image and audio understanding, an optional thinking mode and a 262,144 token context window.
Apertus 1.5 continues pretraining of Apertus 1.0 on 4T extra tokens for the 8B and 2T for the 70B. It adds image and audio understanding through the tokenizer, while the output head stays text only and the architecture is unchanged. Thinking is a chat template flag rather than a separate model, and the paper says it lifts math reasoning by up to 20 points on the 8B. The weights were released on Hugging Face on July 24, 2026.
- Date
- Sunday 11 October 2026
- Lab
- Swiss AI Initiative
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Additional training tokens | 4T (8B), 2T (70B) total 19T and 17T respectively | company |
| Context window | 262,144 tokens four times Apertus 1.0 | company |
| Math reasoning gain with thinking mode | up to 20 points 8B model | company |
The weights came out on July 24, 2026 under Apache 2.0. The paper (Oct 11) and release notes (Oct 15) are the newer documentation.
Sources
This record was checked against its sources on 15 October 2026. How we check