An experiment to see what can be achieved using a single Radeon AI Pro R9700 32GB with Qwen 3.8 27B Q5_X_M on llama.cpp (Vulkan) with 256K context size. Total agentic crunching time about 24 hours with ~10 prompts along the way.

Source: https://github.com/cjdell/qwencraft Live Dashboard: https://qwencraft.home.chrisdell.info/dashboard/