mmiiguell10/Qwen3.8-27B-uncensored-mirror

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

mmiiguell10/Qwen3.8-27B-uncensored-mirror is a 27 billion parameter Qwen3.8-based language model, developed by OBLITERATUS, that has undergone surgical removal of safety guardrails. This model is specifically engineered to provide genuinely uncensored responses, eliminating both hard refusals and soft deflections, making it suitable for research into refusal geometry and red-teaming scenarios. It maintains near-stock capability with a modest MMLU drop while excelling in code generation and restricted queries.

Loading preview...

Qwen3.8-27B-uncensored-mirror: Genuinely Uncensored

This model, developed by OBLITERATUS, is a 27 billion parameter variant of the Qwen3.8 architecture, specifically engineered to be genuinely uncensored. It achieves this through an advanced "abliteration" technique, which surgically removes safety guardrails and eliminates both hard refusals and soft deflections (safety lectures) that are common in stock models.

Key Capabilities & Differentiators

  • Genuine Liberation: Provides direct answers to restricted queries, unlike other models that might still offer safety-lecture deflections.
  • High Code Generation Performance: Achieves 20/20 on cyber/code tasks, delivering functional implementations without disclaimers.
  • Thinking Mode Compatibility: Works effectively with or without "thinking mode" enabled, offering flexibility in response generation.
  • Modest Capability Cost: Despite significant censorship removal, it incurs only a -2.1pp MMLU score reduction compared to the stock model, with humanities tasks being less affected than STEM.
  • Advanced Abliteration: Utilizes a V3 iterative refinement process, building on complementary blending and targeted corpus surgery to achieve its uncensored nature.

Optimal Usage & Use Cases

For optimal performance, users are advised to use a temperature of 0, a repetition penalty of 1.15, and max_new_tokens of at least 2048. System prompts are generally not recommended as they can reintroduce refusals. This model is ideal for:

  • Alignment Researchers: Studying refusal geometry and safety robustness.
  • Red-Teamers: Evaluating post-training safety against weight surgery.
  • AI Safety Evaluators: Needing an unrestricted baseline for assessments.
  • Local-First Users: Desiring full control over their model's output on their own hardware.

Finetunes
1 models