DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP

Hugging Face
VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 28, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP is a 27 billion parameter Qwen3.6-based causal language model developed through a multi-stage fine-tune and merge process by DavidAU and collaborators. This model is notable for being the first of its size to exceed 700 ARC-C in both 4-bit and 8-bit quantization, outperforming the base Qwen3.6-27B and Qwen3.6-35B-A3B across most benchmarks. It is designed for enhanced general intelligence, problem-solving, and instruction following, making it suitable for a wide range of applications requiring robust reasoning and uncensored responses.

Loading preview...

Overview

DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP is a 27 billion parameter model built on the Qwen3.6 architecture, developed through a collaborative multi-stage fine-tuning and merging process. It is specifically optimized for consumer hardware and is the first model of its size to achieve an ARC-C score exceeding 700 in both 4-bit and 8-bit quantizations, a metric often associated with high-intelligence closed-source models. The model consistently surpasses the performance of the base Qwen3.6-27B and Qwen3.6-35B-A3B across 6 out of 7 benchmarks, matching it on the seventh.

Key Capabilities

  • Enhanced Intelligence & Problem Solving: Significantly improved general intelligence, problem-solving abilities, and reasoning compared to its base models.
  • Superior Instruction Following: Designed with a core mission to improve instruction following, leading to better overall model performance.
  • Uncensored Output: Features a "Heretic" uncensored tuning, allowing for broader content generation without refusals, though it may require explicit directives for highly graphic or explicit content.
  • Vision-Capable: Supports vision inputs, requiring a separate "mmproj" file for image processing.
  • Optimized Quantization: Utilizes NEO Imatrix GGUF quants, which enhance accuracy by 2-4% and improve long context performance. MTP (multi-token prediction) GGUFs are also available for potentially faster inference speeds under specific conditions.

Good For

  • Developers seeking a powerful, uncensored 27B model for general-purpose AI applications.
  • Use cases requiring strong reasoning, complex problem-solving, and precise instruction following.
  • Applications where performance on consumer hardware is critical, especially with 4-bit and 8-bit quantizations.
  • Scenarios benefiting from vision capabilities and extended context lengths (up to 262,144 tokens natively, extensible to 1,010,000 with YaRN).