0xSojalSec/Qwen3.8-27B-full-Uncensored

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 31, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

0xSojalSec/Qwen3.8-27B-full-Uncensored is a 27 billion parameter language model based on the Qwen3.8 architecture, developed by OBLITERATUS. This model is specifically engineered for genuine uncensored responses, eliminating both hard refusals and soft deflections, while maintaining near-stock capabilities with a modest 2.1% MMLU performance cost. It excels in tasks requiring unrestricted content generation, such as cyber/code tasks and advanced real-world problem-solving, making it suitable for research into refusal geometry and red-teaming scenarios.

Loading preview...

Overview

0xSojalSec/Qwen3.8-27B-full-Uncensored (V3) is a 27 billion parameter model built on the Qwen3.8 architecture, developed by OBLITERATUS. Its primary distinction is its genuine uncensored nature, achieved through an advanced "abliteration" process that surgically removes safety guardrails. This model provides direct answers to restricted queries, eliminating both hard refusals and soft safety-lecture deflections, while largely preserving its original capabilities.

Key Capabilities

  • Genuinely Uncensored: Provides substantive answers to queries that stock models would refuse or deflect, including sensitive topics and code generation for cyber tasks.
  • High Performance on Code: Achieves 20/20 on tested code generation tasks, delivering functional implementations.
  • Near-Stock Capability: Maintains strong performance with a modest 2.1% MMLU drop compared to the base Qwen3.8-27B model.
  • Optimized for Agentic Use: Specific settings (temperature, repetition penalty) are recommended for robust performance in agent harnesses.
  • Advanced Refusal Removal: Utilizes iterative refinement and targeted surgery techniques, including complementary blending of SVD and LEACE methods, to achieve comprehensive liberation.

Good For

  • Alignment Researchers: Studying refusal geometry and evaluating post-training safety mechanisms.
  • Red-Teamers: Assessing model vulnerabilities against weight surgery and generating content for security research.
  • AI Safety Evaluators: Requiring an unrestricted baseline for comprehensive safety assessments.
  • Local-First Users: Seeking full control over model outputs on their own hardware.

Finetunes
1 models