Qwen models fine tuned to use structured and self-correcting reasoning, primarily from Gemini 3.x models.