Made this game in two prompts with Q4, Qwen 3.8 is amazing

This took one prompt to build, and another follow up prompt to fix two issues (player got stuck with the bomb and broken enemies path-finding), this is only html, css and js, no external assets, all done by Qwen. Using UD-Q4_K_XL in llama.cpp with 128k context and k5_0/v4_1 quantized cache in 2x RTX 3060, Deepseek Harness, around 20 t/s average. The main issue i had wasnt the model but my system RAM (only 32gb), after each context compact i had to restart llama.cpp to release the RAM, but just sending a "continue" put it back on track and it finished the task beautifully.