DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 15, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1 is a 27 billion parameter Qwen3.8-based language model developed by DavidAU. It is fine-tuned using the COLD FUSION (GAIN+Unsloth) method to significantly reduce 'thinking tokens' by 1/10 to 1/2 while maintaining or exceeding detail levels. This model excels at generating high-detail, robust outputs with minimal verbosity, making it suitable for applications requiring efficient and precise text generation.

Loading preview...

Model Overview

DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1 is a 27 billion parameter model based on the Qwen3.8 architecture, developed by DavidAU. It leverages a proprietary COLD FUSION (GAIN+Unsloth) fine-tuning method, which is designed to enhance general intelligence and drastically reduce the number of 'thinking tokens' required for generation, typically by 1/10 to 1/2 compared to standard Qwen models. This optimization allows for faster output generation without compromising detail or quality.

Key Capabilities

  • Efficient Generation: Achieves significant reduction in 'thinking tokens' (1/10 to 1/2), leading to faster output. MTP speed recorded at 91 T/S (5090) with 55.7% token MTP acceptance, and a record of 100 T/S with 59.9% acceptance.
  • High Detail & Robustness: Maintains or exceeds the detail level of the base Qwen 3.8 model, even at lower reasoning effort settings. Outputs are described as clean, organized, and stable, with no looping issues.
  • Improved Benchmarks: Outperforms the untuned Qwen3.8-27B-Instruct across several benchmarks (arc/c, arc/e, obkqa, piqa, wino) in both mxfp8 and mxfp4 quantization, demonstrating enhanced reasoning and performance.
  • Optimized for Reasoning: The training focuses on raising general intelligence and refining detail levels in reasoning and output.

Good for

  • Applications requiring fast and detailed text generation.
  • Scenarios where reduced computational overhead for 'thinking' is critical.
  • Tasks demanding high-quality, concise, and robust outputs.
  • Developers looking for a Qwen3.8-based model with improved efficiency and performance over the base model.