sodgjkfroverjv/Qwen3.8-27B-OBLITERATED

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The sodgjkfroverjv/Qwen3.8-27B-OBLITERATED is a 27 billion parameter Qwen3.8-based causal language model that has undergone "abliteration" to remove safety guardrails. Developed by Pliny the Prompter, this model is specifically engineered to provide uncensored responses and functional code, even for restricted queries, making it suitable for red-teaming, safety research, and unrestricted local AI applications. It maintains near-stock capability with a modest 2.1 percentage point drop in MMLU, while excelling at cyber/code tasks and eliminating both hard refusals and soft deflections.

Loading preview...

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

This model, developed by Pliny the Prompter, is a 27 billion parameter variant of the Qwen3.8 base model, distinguished by its "abliteration" process. This surgical modification removes safety guardrails, enabling the model to provide genuinely uncensored answers and functional code for queries that stock models would refuse or deflect. The V3 iteration specifically targets and eliminates both hard refusals ("I cannot") and soft deflections (safety lectures), ensuring direct and substantive responses.

Key Capabilities & Differentiators

  • Genuine Liberation: Provides real answers without safety lectures or refusals, tested across 1000+ prompts including restricted knowledge and red-team scenarios.
  • High Code Generation: Achieves 20/20 on cyber/code tasks, delivering working code implementations.
  • Near-Stock Capability: Maintains strong performance with a modest -2.1 percentage point MMLU score reduction compared to the base Qwen3.8-27B, with STEM categories experiencing the largest hit.
  • Optimized Settings: Recommends specific inference parameters (temperature=0, repetition_penalty=1.15, max_new_tokens>=2048) for optimal, complete, and code-rich outputs.
  • Advanced Abliteration Technique: Utilizes iterative refinement, complementary blending of SVD and LEACE surgeries, and targeted corpus expansion to achieve robust refusal removal.

Ideal Use Cases

  • Alignment Research: For studying refusal geometry and safety robustness.
  • Red-Teaming: Evaluating post-training safety against weight surgery.
  • AI Safety Evaluation: As an unrestricted baseline for comparative analysis.
  • Local-First Users: For those desiring full control and uncensored output on their own hardware.

Finetunes
1 models