DavidAU/Qwen3.5-9B-Claude-4.6-OS-HERETIC-UNCENSORED-INSTRUCT

Model Overview

DavidAU/Qwen3.5-9B-Claude-4.6-OS-HERETIC-UNCENSORED-INSTRUCT is a 9 billion parameter instruction-tuned model based on the Qwen 3.5 architecture. It was fine-tuned using the Claude-4.6-OS dataset and is explicitly designed to be "HERETIC" and fully uncensored, meaning it will respond to user requests without refusals. The model maintains the strong benchmarks of the original Qwen 3.5 9B while significantly improving instruction following and reducing refusals (6/100 compared to 100/100 for the original).

Key Capabilities

• Uncensored Responses: Designed to follow instructions without ethical or safety refusals.
• Enhanced Instruction Following: Fine-tuned with Claude-4.6-OS datasets for improved adherence to user prompts.
• Multimodal Support: Retains and improves upon the vision capabilities of the base Qwen3.5 model, including image and video input processing.
• Strong Benchmarks: Demonstrates improved performance across various benchmarks, including reasoning (arc, obkqa, wino) and instruction following (boolq, piqa, hswag) compared to the base Qwen3.5-9B and Qwen3.5-9B-Instruct models.
• Long Context Window: Natively supports a context length of 262,144 tokens, extensible up to 1,010,000 tokens using YaRN scaling.

Good for

• Applications requiring direct, unconstrained responses without content filtering.
• Tasks involving multimodal inputs (text, images, video) where instruction adherence is critical.
• Developers seeking a powerful 9B parameter model with strong general language understanding and reasoning, particularly for scenarios where traditional LLM guardrails are undesirable.
• Agentic workflows and tool calling, leveraging the model's robust instruction-following and reasoning capabilities.