I Found a Better Uncensored Model Than Dolphin — Qwen 3.6

An uncensored model built on Qwen 3.6 (35B total, 3B active MoE) — fine-tuned from the censored base into an aggressive uncensored variant with 0 of 465 refusals. In the video I run it with Unsloth Studio and cover the engineering, not the content.

What's covered

• Model specs — Qwen3.6-35B-A3B, multiple quant levels (Q2 → Q8)
• The uncensoring method
• Checking VRAM fit on HuggingFace (the 2-bit quant needs ~11.7GB — won't fit 8GB laptops, so I rent a GPU)
• Unsloth setup, model download, and testing
• Honest verdict vs older uncensored models (Mistral, Dolphin)

Model → https://huggingface.co/HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive · Unsloth Studio → https://unsloth.ai/docs/new/studio · Get RunPod

Full setup + test in the video above. Subscribe for more local-LLM deep dives.