The QwQ reasoning model is significantly leaner than DeepSeek R1 and is said to be more powerful in many areas.