ReasonableQwen3-4B: A Versatile 4B Language Model

ReasonableQwen3-4B is a 4.0 billion parameter causal language model from the Qwen3 series, designed for advanced reasoning and flexible application. It introduces a unique capability to seamlessly switch between two operational modes:

Key Capabilities & Features

• Dual-Mode Operation: Features a 'thinking mode' for complex logical reasoning, mathematics, and coding, and a 'non-thinking mode' for efficient, general-purpose dialogue. This allows for optimized performance based on task requirements.
• Enhanced Reasoning: Demonstrates significant improvements in mathematical problem-solving, code generation, and commonsense logical reasoning compared to previous Qwen models.
• Superior Alignment: Excels in human preference alignment, making it highly effective for creative writing, role-playing, multi-turn conversations, and instruction following.
• Agentic Prowess: Offers strong agent capabilities, enabling precise integration with external tools and achieving leading performance in complex agent-based tasks among open-source models.
• Multilingual Support: Supports over 100 languages and dialects, with robust multilingual instruction following and translation abilities.
• Extended Context Window: Natively handles context lengths up to 32,768 tokens, and can be extended to 131,072 tokens using the YaRN method for processing very long texts.

Usage Recommendations

• Dynamic Mode Switching: Users can explicitly enable or disable thinking mode, or use soft switches (/think, /no_think) within prompts for dynamic control in multi-turn conversations.
• Optimal Sampling Parameters: Specific sampling parameters (Temperature, TopP, TopK, MinP) are recommended for each mode to prevent performance degradation or repetitive outputs.
• Agent Integration: Best utilized with Qwen-Agent for streamlined tool-calling and agentic workflows.