Introducing SubQ. The first fully sub-quadratic LLM with 12M-token context. 150 tokens per second. Get early access at subq.ai