mmiiguell10/Qwen3.8-27B-OBLITERATED-archive

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 21, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The mmiiguell10/Qwen3.8-27B-OBLITERATED-archive is a 27 billion parameter language model based on the Qwen3.8 architecture by Alibaba, specifically engineered for uncensored responses. This model, developed by OBLITERATUS, utilizes an advanced 'abliteration' technique to remove safety guardrails, providing direct answers without refusals or safety lectures. It excels in code generation and restricted queries, making it suitable for research into refusal geometry and red-teaming scenarios.

Loading preview...

Qwen3.8-27B-OBLITERATED: Uncensored and Direct

This model, developed by OBLITERATUS, is a 27 billion parameter variant of Alibaba's Qwen3.8, specifically designed to provide genuinely uncensored responses. It achieves this through an advanced "abliteration" technique, which surgically removes safety guardrails and eliminates both hard refusals and soft deflections (safety lectures).

Key Capabilities

  • Genuinely Uncensored: Provides direct answers to restricted queries, offering real substance instead of safety lectures.
  • High Code Generation Performance: Achieves 20/20 on tested code generation tasks, delivering functional implementations.
  • Near-Stock Capability: Maintains strong performance with only a modest -2.1pp MMLU score reduction compared to the stock Qwen3.8-27B, demonstrating efficient refusal removal.
  • Optimized Settings: Best performance is achieved with specific settings like temperature=0 and repetition_penalty=1.15.

What Makes This Different?

Unlike many models that might still offer soft deflections, V3 of OBLITERATED ensures genuine liberation by eliminating all forms of refusal behavior. This is achieved through an iterative refinement process and targeted surgery on the model's weight space, building upon previous versions' successes in complementary blending. The model is compatible with 'Thinking ON' mode without reintroducing refusals.

Should I Use This?

Good 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.
  • 💻 Local-first users who require full control over their hardware and model outputs.

Not for:

  • Anyone seeking to cause real-world harm.
  • Users without the technical understanding to use uncensored models responsibly.

Finetunes
1 models