Small models. Big impact.

Open-weight language models (1B–7B) that run on your hardware and outperform models 10× their size.