Experimental model; This is a DARE-linear merge of Qwen3-8B instruction finetunes, with rough abliteration vectors extracted and applied to the weights to reduce refusal. For uncensored instruction following, structured output, and basic agentic use.