Axion Labs
Built, trained, and hosted by us
Free to chat with after creating an Axion account. Browser chat does not require an API key.
Lumen 1.2.5
Axion Labs
Fine-tuned Llama 3.1 8B with DPO safety training. Best for general chat, questions, writing, and code explanations. 73% safety pass rate under adversarial testing.
Veil 1.1
Axion Labs
Our first and fastest model — a fine-tuned Llama 3.1 3B for quick general chat.
Research
Safety is the point, not a checkbox
Every model we ship goes through adversarial safety testing before release, and we publish the results.
Lumen 1.2.5 Safety Report
DPO safety alignment results and adversarial test pass rates for our flagship model.
All announcements
Release notes, safety reports, and infrastructure updates as we ship them.
Project Crucible
Our active R&D sprint exploring ternary weights, mixture-of-experts routing, and efficient model training.