QwQ uses inference-time scaling to solve complex reasoning and planning questions, besting OpenAI's o1 in several benchmarks.