Recently, Southeast University officially connected to Alibaba's latest open source model QwQ-32B based on the Shengteng domestic computing power platform, becoming the first university in the country to launch the QwQ-32B service. The dual-track integration of domestic computing power base and large head model will inject stronger AI momentum into teaching and scientific research. QwQ-32B: A powerful replacement for DeepSeek-R1, with performance comparable to top-level models. Although QwQ-32B has only 32B parameters, through large-scale reinforcement learning training, its performance is comparable to DeepSeek-R1 with 671B parameters in tasks such as mathematical reasoning (AIME24 evaluation), code generation (LiveCodeBench), and tool calling (BFCL test), and it even surpasses it in some indicators. The above figure shows the performance comparison of QwQ-32B and other leading models in mathematical reasoning (AIME24 evaluation), code generation (LiveCodeBench), and tool calling (BFCL test). The performance of QwQ-32B is comparable to DeepSeek-R1, and far better than o1-mini and R1 distillation models of the same size.