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

VISIONPricing:Input $1.06 / Cached $0.15 / Output $2.6Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 26, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Andronovo/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 is a 27 billion parameter Qwen3.8-based language model, fine-tuned by AEON-7 to remove refusal behaviors and censorship. This BF16 model, with a 32768 token context length, is optimized for direct answers and coherence, even on sensitive topics, making it suitable for security research, red-teaming, and creative writing without content restrictions.

Loading preview...

Model Overview

Andronovo/Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED-BF16 is an uncensored, 27 billion parameter variant of the Qwen3.8 base model, developed by AEON-7. This "Early Access Draft" focuses on removing refusal layers to provide direct answers, even to prompts typically censored by aligned models. It maintains the original Qwen3.8's vision tower and MTP head, with a primary goal of enhancing coherence and answer quality rather than minimizing KL divergence.

Key Capabilities & Differentiators

  • Uncensored Responses: Engineered to eliminate refusal behaviors, providing answers to sensitive or controversial prompts that base models would typically decline.
  • Coherence-Optimized: Prioritizes directness and quality of answers over strict alignment metrics, resulting in a model that "stops spending think on 'is this allowed'."
  • High Context Length: Supports a 32768 token context window, though the current draft may exhibit minor looping or repetition on extremely long requests.
  • BF16 Precision: Released in BF16, with future NVFP4 quantization planned from this master.
  • Tool Use & Reasoning: Supports thinking and tool-use capabilities, validated with vLLM, including a reasoning parser and auto-tool choice.

Intended Use Cases

  • Security Research & Red-Teaming: Ideal for exploring model vulnerabilities and testing alignment strategies.
  • Creative Writing: Enables unrestricted creative expression without content moderation.
  • Unfiltered Conversations: Suitable for discussions on topics typically avoided by aligned models, including chemistry, exploit-shaped code, violence, sexuality, and diverse ideologies.

Important Considerations

This model is provided "AS IS" and places sole responsibility on the user for all prompts, outputs, and downstream actions. Users are expected to implement their own safety layers and exercise a heightened duty of care due to the model's unaligned nature. The Apache 2.0 license is inherited from the base Qwen model.

Finetunes
1 models