biennequants/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 7, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The biennequants/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU model is a 27 billion parameter Qwen 3.8-based language model developed by DavidAU, fine-tuned for enhanced reasoning and uncensored output. It significantly outperforms base Qwen 3.8 27B models across multiple benchmarks, with ARC-C scores 141 points higher. This model features reduced 'thinking tokens' while maintaining detail, stable 4-bit performance, and is designed for use cases requiring robust analytical capabilities and creative, unrestricted text generation.

Loading preview...

Overview

This model, biennequants/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU, is a 27 billion parameter Qwen 3.8-based language model developed by DavidAU. It is the first release in a series of highly optimized and "heretic'd" (uncensored) versions, built upon extensive fine-tuning stages. The model aims to significantly exceed the performance of all Qwen 27B models, including other fine-tunes, by focusing on enhanced reasoning, reduced token overhead, and uncensored output.

Key Capabilities

  • Superior Benchmark Performance: Achieves an ARC-C score 141 points above the base Qwen 3.8 27B benchmark, with strong improvements across other metrics (ARC-E, BoolQ, HSwag, OBQA, PIQA, Wino).
  • Efficient Reasoning: Features a significant reduction in "thinking tokens" (1/2 to 1/10 of normal Qwen size) while maintaining or increasing detail level, with auto-variable thinking sizes based on prompt.
  • Robust Uncensoring: Undergoes a "heretic'ing" process with a low KLD (0.0025) for very strong decensoring, followed by a precision-engineered "healing" dataset to restore and even boost core metrics post-uncensoring.
  • Stable Quantization: Demonstrates stable performance in 4-bit quantization, retaining approximately 99% of 8-bit performance.
  • Creative & Analytical Depth: Exhibits enhanced depth of thinking and analytical capabilities, with specific versions (e.g., Stage 1b-endgame) showing creative performance upgrades and distinct character.

Good For

  • Advanced Text Generation: Ideal for applications requiring highly detailed, analytical, and creative text outputs without content restrictions.
  • Complex Reasoning Tasks: Excels in scenarios demanding strong logical inference and problem-solving, as evidenced by its high benchmark scores.
  • Resource-Efficient Deployment: Suitable for environments where 4-bit quantized models are preferred, offering near 8-bit performance with reduced memory footprint.
  • Exploratory & Unrestricted Content Creation: Beneficial for developers and users who need a model capable of generating diverse and uncensored content for creative writing, roleplay, or research purposes.

Finetunes
1 models