Model Overview

TeichAI/Qwen3-4B-Thinking-2507-GPT-5.2-High-Reasoning-Distill is a 4 billion parameter Qwen3-based language model developed by TeichAI. This model is a distillation, fine-tuned from unsloth/qwen3-4b-thinking-2507.

Key Capabilities

• High Reasoning: The model's primary strength lies in its enhanced reasoning capabilities, achieved through distillation from 250 examples generated by GPT 5.2 (high reasoning).
• Optimized Training: It was trained 2x faster using Unsloth and Huggingface's TRL library, indicating an efficient fine-tuning process.
• Context Window: Supports a substantial context length of 40960 tokens, allowing for processing and understanding of extensive inputs.

Unique Aspects

This model specifically addresses and fixes formatting issues found in previous GPT 5 distills, aiming for improved data quality in its training examples. Its focus on distilling high-reasoning examples from a powerful source like GPT 5.2 makes it particularly suitable for tasks demanding complex logical thought and problem-solving, distinguishing it from general-purpose language models.