On November 28, 2024, Alibaba’s Qwen team released QwQ-32B-Preview, an experimental 32.5-billion-parameter open-weight model focused on reasoning. The name stands for “Qwen with Questions,” and the team framed it around deliberate, patient thinking: “when given time to ponder, to question, and to reflect, the model’s understanding of mathematics and programming blossoms.” It was one of the first openly downloadable models built explicitly to reason step by step in the style of OpenAI’s then-new o1.
The team reported strong results on hard reasoning benchmarks for the model’s size, including 90.6 percent on MATH-500, 65.2 percent on GPQA, and 50.0 percent on both AIME and LiveCodeBench. They were candid about its preview status, noting that the model “may enter circular reasoning patterns, leading to lengthy responses without a conclusive answer,” along with occasional language mixing.
QwQ-32B-Preview mattered because it brought the reasoning-model paradigm into the open within weeks of o1’s debut. By showing that a mid-sized open model could be taught to reason effectively, it pointed the open ecosystem toward the inference-time-compute approach that DeepSeek-R1 would soon make famous.