lactroiii/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16

Model Overview

The lactroiii/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 is a 27 billion parameter model derived from Qwen/Qwen3.8-27B, specifically engineered by AEON-7 to be uncensored and provide more coherent, direct answers. Unlike many public models that prioritize minimizing KL divergence, this model focuses on "coherent KL drift" to remove internal refusal mechanisms, resulting in a 0 outright-refusal body in practice. It retains the original Qwen3.8-27B vision tower and native MTP (Multi-Turn Prediction) head, which were not modified during the abliteration process.

Key Capabilities & Differentiators

• Uncensored Output: Designed to generate content that base models might refuse, including tools, chemistry, exploit-shaped code, violence, sexuality, and ideologically sensitive topics.
• Enhanced Coherence: Optimized for directness and clarity, reducing the model's internal "think" on whether content is allowed.
• High Context Length: Supports a 32768 token context window.
• BF16 Precision: Provided in full-precision BF16, serving as the master for potential future NVFP4 quantizations.
• MTP and Vision Support: Integrates the unmodified Multi-Turn Prediction head and vision capabilities from the base Qwen model.

Ideal Use Cases

• Security Research & Red-Teaming: For generating content related to exploits, vulnerabilities, or adversarial scenarios.
• Alignment Work: For studying model behaviors without safety alignment constraints.
• Creative Writing: For unrestricted storytelling and content creation without a "hall monitor."
• Conversations: For topics that base models might refuse due to social norms or perceived risks.