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.