Independent lab,
open by default.

Empero is an independent AI research lab. We build efficient language models you can own and run yourself — and the full open pipeline behind them. Our published flagship, Qwythos-9B-Claude-Mythos-5-1M, is a Claude Mythos 5 distill with a 1M-token context; our in-house model, Claire, is a 6B-A500M mixture-of-experts trained from scratch on a custom architecture and recipe. Building a better tomorrow, in the open.

We own the stack end to end: microverse finds better architectures, SFTSuite builds the curriculum data, and the model is trained on hardware we can reach. Weights, code and datasets ship in the open — no proprietary kernels, no waitlists, inspectable defaults. Built in Germany; the work is the work.