A user successfully generated a one-page self-portrait website using the Qwen3.8-Flash-Next model via llama.cpp on an RTX 3060. Despite the model exceeding available RAM, the system utilized page caching to process 269K tokens over nine hours. Interestingly, the model signed the final output as "Claude."