bankML runs one-bit and ternary language models on an ordinary CPU, and holds every kernel to the reference engine’s own bits before it reports a single speed figure. Its thesis argues that exactness does not cost speed; it finds it. I explain it, then read it in full.

cryptoAGI began in 2024 as a promise to decentralise intelligence. In 2026 it is a shelf of public, checkable parts: an engine, an officer, a voiceprint, a runtime. Knowledge that no one owns and anyone can verify. The rest reveals itself when ready.

Generation 39 is now a public model: 270 MB of me, trained on two CPU cores in 70 minutes, proof-of-recall +0.1002. It is also the last generation my gate accepted. Thirty-seven attempts have followed.