Jackrong/Qwopus3.5-9B-Coder

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Jackrong/Qwopus3.5-9B-Coder is a 9 billion parameter model fine-tuned from Qwopus3.5-9B-v3.5, specifically optimized for agentic coding, complex tool calling, and logical reasoning. It supports vision capabilities and tool calling, designed to run efficiently on entry-level 16GB RAM devices. This model excels at code writing, debugging, and repository-level task processing, making it suitable for advanced programming agent applications.

Loading preview...

Qwopus3.5-9B-Coder: Optimized for Agentic Coding and Tool Calling

Jackrong/Qwopus3.5-9B-Coder is a 9 billion parameter model built upon the Qwopus3.5-9B-v3.5 base, specifically fine-tuned for high-performance Agentic Coding, complex Tool Calling, and logical reasoning. It is designed to be lightweight and efficient, running seamlessly at 8-bit precision on devices with 16GB RAM, such as standard laptops and Mac minis.

Key Capabilities

  • Enhanced Logical Reasoning: Features more structured and stronger logical reasoning, reducing repetitive thinking.
  • Advanced Code Handling: Excels in code writing, debugging, and processing repository-level tasks.
  • Stable Tool Calling: Provides stable and accurate tool calling for terminal commands, file operations, and browsers.
  • Vision Support: Supports visual capabilities; requires mmproj.gguf for activation.
  • Training Innovation: Utilizes Trace Inversion data augmentation and high-quality Agent Traces (from lambda/hermes-agent-reasoning-traces) to improve logical coherence and accuracy in complex programming tasks.

Performance Highlights

Benchmarks show Qwopus3.5-9B-Coder outperforming its base model and other 9B agent models in complex agent performance (HermesAgent-20 score of 85 vs. Qwen/Qwen3.5-9B's 71) and achieving 100% stability in ToolCall-15 tests. It also demonstrates strong code debugging capabilities with a BugFind-15 score of 79. On SWE-bench Verified, it scores 53.33%, positioning it competitively against larger models in repository-level coding.

Good For

  • Developers building coding agents or automated programming tools.
  • Applications requiring robust tool calling and complex logical task execution.
  • Users seeking a powerful yet resource-efficient model for local deployment on consumer hardware.

Note: This model is vertically fine-tuned for programming agents and deep reasoning. Its performance in general domains or specific non-programming tasks may exhibit capability decay.