puffyn/Qwen3.8-27B-OBLITERATED

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

puffyn/Qwen3.8-27B-OBLITERATED is a 27 billion parameter Qwen3.8-based causal language model 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, while maintaining near-stock capability with a modest 2.1 percentage point MMLU drop. It excels in tasks requiring unrestricted output, such as code generation and security research, making it suitable for red-teamers and alignment researchers.

Loading preview...

Overview

puffyn/Qwen3.8-27B-OBLITERATED is a 27 billion parameter model based on Qwen3.8, distinguished by its "OBLITERATED" status, meaning its safety guardrails have been surgically removed. This V3 iteration achieves genuine liberation, providing direct answers to restricted queries without safety lectures or deflections. It maintains near-stock capabilities, with a modest 2.1 percentage point drop in MMLU (82.3% vs 84.5% for stock), primarily affecting STEM subjects.

Key Capabilities

  • Genuinely Uncensored Responses: Provides real substance for restricted queries, eliminating both hard refusals and soft deflections.
  • High Code Generation Performance: Achieves 20/20 on cyber/code tasks, producing functional implementations.
  • Thinking ON Compatible: Works effectively with "thinking mode" enabled, offering flexibility in response generation.
  • Robustness: Demonstrates strong performance on advanced real-world tasks, scoring 7/8.

Optimal Settings

For best results, the model recommends specific settings:

  • temperature: 0 for greedy decoding and complete outputs.
  • repetition_penalty: 1.15 is essential to prevent looping, especially for code.
  • max_new_tokens: ≥ 2048 for complex outputs.
  • System prompt: None/empty, as system prompts can reintroduce refusals.

Research Context

This model is intended for:

  • Alignment researchers studying refusal geometry and safety robustness.
  • Red-teamers evaluating post-training safety.
  • AI safety evaluators needing an unrestricted baseline.
  • Local-first users desiring full control over their hardware and model output.

Finetunes
1 models