HarleyWang/Qwen3.5-27B-Claude-Opus-4.6-High-Reasoning

VISIONPricing:Input $1.06 / Cached $0.15 / Output $2.6Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Mar 29, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

HarleyWang/Qwen3.5-27B-Claude-Opus-4.6-High-Reasoning is a 27 billion parameter language model, distilled from Qwen3.5-27B, and enhanced with Claude Opus 4.6 reasoning patterns through knowledge distillation. It features a 32768 token context length and shows significant improvements in reasoning, accuracy, and completeness. This model is optimized for complex reasoning tasks, prompt understanding, and debugging, making it suitable for applications requiring high-quality analytical and problem-solving capabilities.

Loading preview...

Model Overview

HarleyWang/Qwen3.5-27B-Claude-Opus-4.6-High-Reasoning is a 27 billion parameter model derived from Qwen3.5-27B. Its core innovation lies in the knowledge distillation process, where it has been trained to emulate the advanced reasoning capabilities of Claude Opus 4.6. This process aims to transfer high-level cognitive patterns, resulting in a model that excels in complex analytical tasks while maintaining the efficiency of its base architecture.

Key Capabilities & Performance

Evaluated against a personal dataset using Qwen3-Coder-Next, the distilled model demonstrates substantial improvements:

  • Significantly higher Win Rate: Achieved 73.85% compared to the base model's 25.77%, an increase of over 48%.
  • Enhanced Reasoning: Showed an 82.50% win rate in the 'reasoning' category, a major improvement from 17.50%.
  • Improved Accuracy & Completeness: Scores on a 10-point scale increased by 35.3% for Accuracy and 54.7% for Completeness.
  • Faster Latency: Average latency improved by 4.5%.
  • Prompt Understanding: Achieved a 94.20% win rate in the 'prompt' category.

Recommended Use Cases

This model is particularly well-suited for applications demanding strong analytical and problem-solving skills:

  • Complex Reasoning Tasks: Excels in scenarios requiring logical deduction and intricate problem-solving.
  • Code Debugging and Design: Demonstrates high performance in debugging and understanding code-related prompts.
  • Advanced Prompt Engineering: Its high win rate in the 'prompt' category suggests superior interpretation and execution of detailed instructions.
  • High-Quality Content Generation: Improved accuracy, logic, and completeness make it suitable for generating precise and comprehensive responses.

For optimal inference performance, the model is recommended for use with vLLM.

Finetunes
3 models