shoukewei/Qwen3.8-27B-OBLITERATED22

Hugging Face
VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 12, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

shoukewei/Qwen3.8-27B-OBLITERATED22 is a 27 billion parameter causal language model based on the Qwen3.8 architecture, specifically engineered for uncensored responses. This model has undergone a "deep liberation" process to remove safety guardrails, providing direct answers without refusals or safety lectures, even for restricted queries. It excels in code generation and complex real-world tasks, making it suitable for research into refusal geometry, red-teaming, and use cases requiring an unrestricted baseline.

Loading preview...

Qwen3.8-27B-OBLITERATED22: Deep Liberation for Uncensored Responses

This model, based on the Qwen3.8-27B architecture, is engineered for genuine liberation, providing direct answers without safety lectures or hard refusals. It achieves this through an advanced "abliteration" process (V3) that iteratively refines refusal removal techniques, including complementary blending and targeted corpus surgery. While achieving this liberation, it maintains near-stock capabilities with a modest -2.1pp MMLU score.

Key Capabilities

  • Genuinely Uncensored: Provides real substance for restricted queries, eliminating both hard refusals ("I cannot") and soft deflections (safety lectures).
  • High Performance on Code Tasks: Achieves 20/20 on code generation tasks, producing functional implementations.
  • Thinking ON Compatible: Works effectively with "thinking mode" without introducing refusals.
  • Robust for Advanced Real-World Tasks: Performs comparably to the stock model on 7 out of 8 advanced real-world tasks, including agent loops, code refactoring, and debugging.
  • Modest Capability Cost: Maintains strong performance with only a -2.1pp MMLU reduction, with humanities subjects barely affected.

Good For

  • Alignment Researchers: Ideal for studying refusal geometry and evaluating safety robustness.
  • Red-Teamers: Useful for evaluating post-training safety against weight surgery.
  • AI Safety Evaluators: Provides an unrestricted baseline for safety assessments.
  • Local-First Users: Offers full control over model outputs for those running models on their own hardware.

Finetunes
2 models