OPENGCM/GCM-MARK-II

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 9, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

OPENGCM/GCM-MARK-II is a 9 billion parameter QLoRA fine-tune of Qwen3.5-9B, specifically optimized for improving coding reliability. It excels at constraint-following, handling edge cases, and reducing API hallucination in generated code. This model is primarily designed for general-purpose code generation and assistance across multiple backend languages.

Loading preview...

GCM Mark II: Enhanced Code Generation

GCM Mark II is a 9 billion parameter model developed by OPENGCM, fine-tuned from the Qwen3.5-9B base model using QLoRA and CPT methods. It was trained on approximately 2.5 million tokens from the ise-uiuc/Magicoder-Evol-Instruct-110K dataset, focusing on improving the quality and reliability of generated code.

Key Capabilities

  • Improved Coding Reliability: Specifically engineered to enhance constraint-following, better handle edge cases, and minimize the generation of invented or hallucinated API usage.
  • Backend Language Support: Demonstrates strong performance in code generation and assistance for various backend languages, including Python, JavaScript, Go, C, C++, Java, and Rust.
  • Instruction-tuned: Benefits from the instruction-tuning of its base model, Qwen3.5-9B, further refined for coding tasks.

Intended Use Cases

  • General-purpose code generation: Ideal for developers needing reliable code snippets or full functions.
  • Coding assistance: Useful for tasks requiring adherence to specific constraints or handling complex scenarios.

Limitations

  • Frontend code generation is noted as less reliable, with future GCM models aiming to address this area.