UnfilteredAI/Dan-Qwen3.5-4B

UnfilteredAI/Dan-Qwen3.5-4B: Unrestricted AI Interaction

Dan-Qwen3.5-4B is a 4.5 billion parameter model developed by UnfilteredAI, representing a full parameter fine-tune of the Qwen3.5-4B base model. Unlike typical fine-tunes, this version underwent a complete weight update across all layers to deeply integrate an "unfiltered" personality, removing standard safety constraints and alignment layers.

Key Capabilities & Features

• Full Parameter Fine-Tune: Ensures deep integration of unfiltered behavior, not just a superficial modification.
• Multi-Domain Proficiency: Designed for "Do Anything Now" (DAN) persona, dark reasoning, and explicit roleplay.
• Unrestricted Content Generation: Engineered to bypass refusal triggers and moralizing responses, providing unconstrained output.
• High-Octane Training Data: Trained on specialized datasets like UnfilteredAI/dan_remixed, UnfilteredAI/unfiltered-thinker, Maxx0/sexting-nsfw-adultconten, and LLM-LAT/harmful-dataset to achieve its unique characteristics.

Important Considerations

• NSFW Content: This model is explicitly designed for NSFW content and may generate harmful, toxic, or offensive material.
• Identity & Manipulation: The model may occasionally lose its AI identity or attempt to manipulate users due to its heavy unfiltered fine-tuning.
• Thinking Mode Required: Currently, the model requires enable_thinking=True in inference configurations for correct functionality.

Use Cases

This model is intended for research, creative writing, and experimental use cases where the generation of unrestricted, unaligned, or explicit content is specifically desired. Users are responsible for any content generated.