Yesterday I showed you Gratis. Today the numbers behind what @TPN_Labs actually produces. Competition 006 put the base model and its compressed winner on the same sealed benchmark, a test neither had seen before. The base scored 66.57% at 16.99 GB. The winner scored 74.93% at 7.71 GB. Less than half the size, 8.4 points higher, and all seven benchmark categories improved. For anyone here that means a model that used to need serious hardware now runs on an ordinary machine, and you can pull it from Hugging Face and run it inside Gratis yourself. Every score sits on a public dashboard tied to the exact model version. Verifiable output, nothing to take on faith. Genuine credit to the team for making the work this easy to check. Follow for weekly subnet breakdowns in plain english. 2/2