miss-t/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16
The miss-t/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 is a 27 billion parameter Qwen3.8-based model, developed by AEON-7, that has been specifically 'abliterated' to remove safety alignments and refusal behaviors. This uncensored BF16 model prioritizes coherence and direct answers over minimizing KL divergence or avoiding judge-R flags, making it suitable for use cases requiring unrestricted content generation. It retains the original Qwen3.8 vision tower and native MTP head, offering enhanced capabilities for security research, red-teaming, and creative writing without content restrictions.
Loading preview...
Model Overview
The miss-t/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 is a 27 billion parameter model derived from Qwen/Qwen3.8-27B, meticulously modified by AEON-7. This version is specifically engineered to be uncensored and unaligned, meaning it has undergone a process called "abliteration" to remove typical safety alignments and refusal behaviors often found in base models. The development prioritized coherence and direct answers over strict adherence to low KL divergence or avoiding judge-R flags, which are often associated with over-editing models.
Key Characteristics
- Uncensored & Unaligned: Designed to provide direct answers without internal refusal mechanisms, even for sensitive or controversial prompts.
- Coherence-Optimized: Abliteration focused on improving answer quality and directness, rather than simply minimizing KL drift.
- Zero Outright Refusal: In practice, the model exhibits zero outright refusals, providing content that base models might typically decline.
- Retains Base Capabilities: The original Qwen3.8 vision tower and native MTP (Multi-Turn Prediction) head remain untouched and fully functional.
- BF16 Precision: This release is the full-precision BF16 reference, with potential future NVFP4 versions derived from it.
Intended Use Cases
This model is particularly well-suited for applications where unrestricted content generation and direct responses are critical, including:
- Security Research & Red-Teaming: For exploring vulnerabilities and generating content without artificial constraints.
- Alignment Work: For studying and understanding model behaviors without safety layers.
- Creative Writing: For generating diverse and unrestricted narratives, dialogue, and scripts.
- Conversations: For engaging in discussions that aligned models might refuse due to social norms or content policies.
Users are explicitly reminded of their sole responsibility for all outputs and downstream actions, as the model does not impose internal safety layers. Appropriate external safety measures are strongly recommended for any production deployment.