I hadn’t realized they updated this model last month! You can use it here.
Apertus 1.5 70B is Switzerland’s latest fully open multilingual large language model developed by the Swiss AI Initiative. It is designed to advance transparent, compliant, and high-performance AI while remaining fully open throughout the entire training pipeline.
- Swiss AI Initiative : Developed by the Swiss AI Initiative to provide a fully open, transparent, and compliant foundation model.
- Fully Open : Open model weights, open training data, complete training recipes, and publicly documented training methodology.
- Massively Multilingual – Native support for 1,811 languages, enabling broad multilingual understanding and generation.
- Compliance First – Trained using fully compliant datasets while respecting retrospective opt-out requests and reducing memorization through the Goldfish loss methodology.
- Long Context – Supports context windows of up to 262,144 tokens, enabling large document analysis and extended conversations.
- Tool Calling Support – Supports agentic workflows and tool use when deployed with compatible inference frameworks such as recent versions of vLLM.
- Open Deployment Ecosystem – Compatible with Transformers, vLLM, SGLang, MLX and GGUF deployment formats.
I’ve been trying it this evening. A bit more sycophancy than I’d like, and it has made technical mistakes in area’s I’m an expert. But nothing egregious when web-search was enabled, and I believe the hosted version at publicai doesn’t have thinking enabled (which makes a big difference in computational tasks).
That said, it’s not far off from models ~6 months ago imo. Which seems a fine gap for the ethical/compliant/eco version.
Why not just use your own brain for tasks in those areas where you are an expert? Honest question. Don’t you enjoy those tasks if you’re an expert?
That’s how you test stuff, look for mistakes
But they’re an expert. Why the need for someone/something else to do their job? (And risk discovering their job safety is at risk.)
If i was to evaluate something, i‘d probably test it on something i’m an expert in, because then i can immediately spot the mistakes. If i test it one something I really don’t know at all, then fact checking/ finding mistakes is gonna become tedious.
So, these cases exist:
Sorry if this is all obvious, I just don’t see the point.
I am making a public recommendation on the internet. I am good at some things that AI will often be used for, such as complicated mathematics. So, as a reviewer, I want to see if the AI will help people with those things well (specifically, help people understand complicated math papers).
Gotcha
As another example, teachers pay teachers to share their created teaching materials. Big part of their job; huge need to share results/progress/ideas.
No, not all tasks that require expertise are fun. For example, one must be a very talented plumber to properly manage a backed-up city sewer. But I suspect the plumber doesn’t enjoy a lot of the tasks involved.
In my setting, technical writing of obvious (to me) facts is not a great time. First drafts are murder for me to create.
Understandable, I suppose.
How did you become an expert in something that you absolutely hate doing?
Because I really like parts of it! Solving puzzles is a grand time. Getting to see students go from ignorance to understanding is quite rewarding.
But writing that first draft. Writing the AI can have.
Kinda like how one often commutes to rewarding work, but hates the commute.
Thanks, this paints a clearer picture 🙂