Notes from building AI in production.
How an open model hit true 1M context — and 7/8 agentic eval — on two RTX 3090s.
We use privacy-friendly analytics to see how the site is used. Nothing loads until you choose. You can change your mind anytime.