Jackrong/Qwen3.5-9B-GLM5.1-Distill-v1

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 15, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Cold

Jackrong/Qwen3.5-9B-GLM5.1-Distill-v1 is a 9 billion parameter Qwen3.5-based model, distilled from GLM-5.1 reasoning data using Supervised Fine-Tuning (SFT) with Unsloth. This model is specifically optimized to enhance structured reasoning ability, improve instruction-following consistency, and activate latent knowledge through better reasoning structures. It excels in tasks requiring transparent internal logic, such as offline analytical tasks, coding, and mathematics, by learning a task-first and structure-driven reasoning pattern.

Loading preview...

Model Overview

Jackrong/Qwen3.5-9B-GLM5.1-Distill-v1 is a 9 billion parameter model built upon the Qwen3.5-9B base, fine-tuned using Supervised Fine-Tuning (SFT) and distillation techniques with Unsloth. Its core innovation lies in distilling high-quality reasoning data from a GLM-5.1 teacher model, specifically the Jackrong/GLM-5.1-Reasoning-1M-Cleaned dataset. This process aims to transfer a robust reasoning structure and problem-solving style, rather than merely imitating outputs.

Key Capabilities & Improvements

  • Enhanced Structured Reasoning: The model learns a task-first, structure-driven reasoning scaffold, emphasizing task decomposition, constraint extraction, and step-by-step logic, distinct from previous Claude-style scaffolds.
  • Improved Instruction Following: Demonstrates better consistency in adhering to instructions, leading to more stable and predictable outputs.
  • Clearer Output Organization: Produces more structured and readable responses, particularly beneficial for complex problems.
  • Domain-Aware Reasoning: Benefits from a training dataset with wide domain coverage and strong problem decomposition patterns, supporting multilingual reasoning.

Intended Use Cases

  • Complex Problem Solving: Designed for tasks requiring multi-step reasoning, such as coding, mathematics, and STEM problems.
  • Analytical Tasks: Ideal for offline analytical tasks where transparent internal logic and structured outputs are crucial.
  • Learning & Research: Serves as an experimental model for academic research and technical exploration, focusing on the transfer of reasoning procedures through distillation.

While gains are incremental for a 9B model, the primary benefit is improved stability, structure, and consistency in complex reasoning tasks, rather than a dramatic jump in raw capability. The distillation philosophy focuses on enabling the student model to better utilize and activate its existing knowledge through structured reasoning.