hadadxyz/OpenSonnet-Lite

OpenSonnet-Lite: Compact Reasoning for Everyday Use

OpenSonnet-Lite, developed by hadadxyz, is a 4.0 billion parameter causal language model fine-tuned from Qwen/Qwen3-4B-Thinking-2507. It is engineered to deliver robust Chain-of-Thought (CoT) reasoning, aiming for performance comparable to commercial models like Claude Sonnet 4.6, while remaining fully open-weights and accessible. A key differentiator is its restoration of multi-turn reasoning, addressing limitations in the base model's chat template to ensure consistent and coherent dialogue over extended conversations.

Key Capabilities

• Enhanced Chain-of-Thought Reasoning: Approaches the reasoning performance of Claude Sonnet 4.6, even on limited hardware.
• Multi-Turn Coherence: Corrected chat template enables consistent reasoning across long, multi-turn dialogues.
• Large Context Window: Features a native context length of 262,144 tokens, supporting extensive input and output.
• Optimized for Efficiency: Designed for everyday use, even on hardware with limited resources.

Good For

• Applications requiring strong, consistent reasoning in conversational AI.
• Deployments on hardware with moderate resource constraints.
• Complex tasks benefiting from Chain-of-Thought prompting and long context.
• Developers seeking an open-weights model with advanced reasoning capabilities for research and general-purpose use.