Cclilqwen: A Specialized 0.6B Model for Creative, Code, and Agentic Tasks

Cclilqwen is a compact yet powerful 0.6 billion parameter model built upon the Qwen3-0.6B base, developed by Asystemoffields. It stands out due to its unique training methodology, which involved fine-tuning four specialist models (for math self-play, creative writing, Python coding, and tool use) and then merging them using a manual DARE-TIES approach. This process resulted in a highly capable model optimized for a diverse set of tasks.

Key Capabilities

• Enhanced Math Reasoning: Achieves 48.5% on GSM8K, nearly doubling the base Qwen3-0.6B's performance.
• Robust Tool Calling: Demonstrates a 100% success rate in generating valid JSON and correct <tool_call> tags for agentic workflows.
• Proficient Code Generation: Successfully generates correct Python solutions for common programming problems like palindrome, fibonacci, and stack implementations.
• Versatile Creative Writing: Capable of producing vivid prose, including gothic horror and dark humor, with effective perspective shifts.

Good for

• Applications requiring a small, efficient model with strong multi-modal capabilities.
• Developers needing reliable tool-calling functionality for agentic systems.
• Generating creative text or code snippets where a compact model is preferred.
• Use cases where a balance of mathematical reasoning, coding, and creative output is essential, despite the parameter size.