Apertus 1.5 technical report is out!
Ever felt like you wanted to know more about what goes into making modern multimodal LLMs?
The team behind the free, copyright and privacy - compliant, heavily multilingual and multicultural LLM has 134 pages + code + data + artifact + intermediate checkpoints for you!
Including, but not limited to:
- How exactly we defined security and
- What Apertus is aligned for?
- How much CO2 we emitted?
https://arxiv.org/pdf/2610.15542
1/2
And as Apertus' safety and security coordinator, I am particularly proud of our transparency in how we defined risks, how we evaluated them, and how they map to common Gen AI risk management frameworks.
We aren't just saying that our models are safe; we are saying how we evaluated what risks before making a decision to release them, including the parts we wish we could have tested better.
And if you find something we forgot: safety@apertus-ai.org is opened and monitored.
2/2