DavidAU/Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored
DavidAU/Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored is a 27 billion parameter Qwen 3.8-based language model developed by DavidAU, featuring significantly reduced 'thinking tokens' and enhanced intelligence. It incorporates five distinct thinking and five instruct modes, switchable on the fly, and excels in instruction following, attention to detail, and consistent multi-stage generations. This model is a 'Light to Moderate Heretic/uncensored' version, balancing performance with a lower refusal rate compared to untuned Qwen 3.8 models.
Loading preview...
DavidAU/Qwen3.8-27B-TWIN-TURBO-Fable-Cold-Fusion-709-L-Uncensored
This 27 billion parameter model, built on the Qwen 3.8 architecture by DavidAU, introduces TWIN-TURBO technology, drastically reducing the number of 'thinking tokens' required (from 1/2 to as low as 1/20) while boosting overall intelligence and performance. It is designed for superior instruction following, meticulous attention to detail, and consistent, high-quality generations, particularly excelling in deep detail and multi-stage drafting tasks.
Key Capabilities & Features
- Reduced Thinking Tokens: Achieves higher performance with significantly fewer internal processing tokens.
- Dynamic Modes: Offers 5 distinct thinking modes and 5 instruct modes, switchable via API or directly within chat messages.
- Enhanced Instruction Following: Demonstrates superior and consistent adherence to instructions.
- Detail-Oriented Generations: Excels in tasks requiring deep detail, double-checking, and multi-stage drafting.
- 'Light to Moderate Heretic' Version: Features a refusal rate of 68/100, offering a more balanced approach compared to untuned models (86/100) while maintaining strong performance.
Performance Highlights
Benchmarks (mxfp8) show this model outperforming base Qwen 3.8-27B across various reasoning tasks:
- ARC-C: 0.709 (vs 0.591 for base Qwen3.8-27B)
- ARC-E: 0.876 (vs 0.782 for base Qwen3.8-27B)
- BoolQ: 0.914 (vs 0.896 for base Qwen3.8-27B)
These metrics indicate a substantial improvement in reasoning and general intelligence. The model's 'thinking mode' is noted to exceed 'instruct mode' benchmark scores in most real-world applications.
Should I use this for my use case?
This model is ideal for applications requiring precise instruction following, detailed and consistent output, and efficient processing due to its reduced thinking token usage. Its 'Light to Moderate Heretic' nature makes it suitable for use cases where a balance between performance and content moderation is desired, without the strict refusal rates of fully aligned models. Consider this model if your application benefits from advanced reasoning, dynamic operational modes, and a more unconstrained generation capability than standard Qwen 3.8 models.