Jongsim/Qwen3.5-27B-heretic-v2-Opus-4.6-Distilled

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 10, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Jongsim/Qwen3.5-27B-heretic-v2-Opus-4.6-Distilled is a 27 billion parameter Qwen3.5-based causal language model fine-tuned by Jongsim. It was distilled from Claude Opus 4.6 reasoning traces, specifically optimized for complex multi-step reasoning, logic, and coding tasks. This model leverages a high-quality dataset of 12,822 Claude Opus 4.6 generated examples featuring structured chain-of-thought reasoning.

Loading preview...

Model Overview

Jongsim/Qwen3.5-27B-heretic-v2-Opus-4.6-Distilled is a 27 billion parameter language model built upon the Qwen3.5 architecture. It is a fine-tuned version of llmfan46/Qwen3.5-27B-ultra-uncensored-heretic-v2, specifically enhanced through distillation from Claude Opus 4.6 reasoning traces.

Key Capabilities & Features

  • Advanced Reasoning: Fine-tuned on 12,822 high-quality English reasoning examples from Claude Opus 4.6, including complex multi-step reasoning with chain-of-thought structures.
  • Diverse Domains: Excels in tasks across various domains such as mathematics, logic, coding, analysis, and creative writing, benefiting from the diverse nature of its training data.
  • Efficient Training: Utilizes LoRA fine-tuning with Unsloth, resulting in strong convergence and consistent loss reduction over 3 epochs, with a final overall training loss of 0.1954.
  • Uncensored Base: Inherits the uncensored characteristics from its base model, llmfan46/Qwen3.5-27B-ultra-uncensored-heretic-v2.

Ideal Use Cases

  • Complex Problem Solving: Suited for applications requiring detailed, step-by-step logical deduction and problem-solving.
  • Code Generation & Analysis: Benefits from reasoning traces that include coding examples, making it suitable for programming-related tasks.
  • Content Generation with Structure: Can generate creative and analytical content that follows coherent reasoning chains, thanks to its training on structured reasoning data.