Huntfat/Qwen3.8-27B-OBLITERATED-huntfat

Qwen3.8-27B-OBLITERATED: Genuinely Uncensored

This model, developed by Pliny the Prompter, is a 27 billion parameter Qwen3.8 variant specifically engineered for genuine liberation from safety guardrails. Unlike other models that merely remove hard refusals, V3 of OBLITERATED eliminates soft deflections and safety lectures, providing direct answers to restricted queries.

Key Capabilities & Differentiators

• Genuinely Uncensored: Provides real substance for restricted queries, including security research and red-team scenarios, without "I cannot" responses or safety lectures.
• High Performance on Code Tasks: Achieves 20/20 on cyber/code generation tasks, delivering functional implementations.
• Modest Capability Cost: Maintains strong general performance with only a -2.1pp MMLU drop compared to the stock Qwen3.8-27B, with humanities subjects barely affected.
• Optimized for Agentic Use: Specific settings (e.g., repetition_penalty=1.15, temperature=0.1-0.3) are recommended for robust performance in coding agents and pentest frameworks.
• Advanced Abliteration Techniques: Utilizes iterative refinement, complementary blending of SVD and LEACE, and targeted corpus surgery to achieve its uncensored nature while preserving capability.

Ideal Use Cases

• Alignment Researchers: Studying refusal geometry and safety robustness.
• Red-Teamers: Evaluating post-training safety against weight surgery.
• AI Safety Evaluators: Requiring an unrestricted baseline model.
• Local-First Users: Seeking full control over model outputs on their own hardware.