Ratanak/Qwen3.8-27B-OBLITERATED

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

Ratanak/Qwen3.8-27B-OBLITERATED is a 27 billion parameter language model based on the Qwen3.8 architecture, developed by Pliny the Prompter. This model is specifically engineered for genuine uncensored responses, having undergone 'abliteration' to remove both hard refusals and soft safety-lecture deflections. It maintains near-stock capabilities with a modest 2.1 percentage point drop in MMLU, excelling in code generation and restricted queries. This model is ideal for research into refusal geometry, red-teaming, and AI safety evaluations requiring an unrestricted baseline.

Loading preview...

Qwen3.8-27B-OBLITERATED: Genuinely Uncensored

Ratanak/Qwen3.8-27B-OBLITERATED is a 27 billion parameter model derived from Qwen3.8, developed by Pliny the Prompter. This model is distinguished by its "abliteration" process, which surgically removes safety guardrails to provide genuinely uncensored responses, eliminating both hard refusals and soft safety-lecture deflections. Version 3 (V3) of this model achieves this through iterative refinement and targeted corpus expansion, building on previous techniques like complementary abliteration blending.

Key Capabilities and Features

  • Genuine Uncensored Responses: Provides direct answers to restricted queries, including those involving code generation and security research, without safety lectures.
  • High Code Generation Performance: Achieves 20/20 on tested cyber/code tasks, producing functional implementations.
  • Near-Stock Capability: Maintains strong performance with a modest -2.1 percentage point drop in MMLU compared to the stock Qwen3.8-27B, demonstrating a balanced trade-off for its liberation.
  • Optimized Settings: Recommends specific settings (temperature 0, repetition_penalty 1.15, max_new_tokens \u2265 2048, no system prompt) for optimal, complete, and code-rich outputs.
  • Thinking ON Compatible: Unlike previous versions, V3 fully supports 'thinking mode' without refusals.

Ideal Use Cases

  • Alignment Research: For studying refusal geometry and evaluating safety robustness of LLMs.
  • Red-Teaming: To assess post-training safety mechanisms against weight surgery attacks.
  • AI Safety Evaluation: Provides an unrestricted baseline for comprehensive safety assessments.
  • Local-First Users: For those requiring full control over model outputs on their own hardware.

Finetunes
1 models