Qwen3.5 GGUF merge lands, quality slips

Interesting as a reproducible local-LLM workflow, but this reads more like a cautionary tale about quantization loss than a clean model recommendation.

• –It mixes two well-known community forks: HauhauCS's uncensored Qwen3.5-35B-A3B and samuelcardillo's Claude 4.6 Opus reasoning distillation.
• –The Colab script is the real utility here, because it packages big-GGUF merging and quantization into something hobbyists can run without workstation-class storage.
• –The author's own quality warning matters more than the release itself: the Q4_0 pass appears to shed enough information to make the result worse than the source quants.
• –For deployers, the takeaway is to test higher-fidelity outputs first and treat aggressive quantization as a space-saving compromise, not a free win.