infotech

Alibaba Has Launched The AI Model QwQ-32B, Which Is Claimed To Surpass Both OpenAI And DeepSeek

The technology company Alibaba has launched a reasoning model named QwQ-32B. This reasoning model represents a new category of AI models, designed to address complex problems through a series of simple and manageable steps, utilizing logical reasoning to arrive at solutions. The AI model simulates problem-solving in a manner akin to human logical conclusions. Although QwQ-32B was initially available in preview mode in November 2024, it has now been officially released in March 2025. 

QwQ-32B is designed as an affordable open-source AI model, featuring 32 billion parameters. In AI models, parameters are the internal variables learned during training. Generally, models with a higher number of parameters tend to be more accurate and capable of handling more complex tasks compared to those with fewer parameters. "Alibaba Cloud introduces QwQ-32B, a compact reasoning model with only 32 billion parameters. This model will deliver performance comparable to other advanced models with larger parameter counts," stated Alibaba, as reported by KompasTekno from the company's official blog on Sunday, March 9, 2025. 

While Alibaba did not specify the other models referenced, it included graphics comparing the performance of QwQ-32B against several other reasoning models, including DeepSeek R1 and OpenAI's o1-mini, across various tests. Notably, DeepSeek R1 is a reasoning model with 671 billion parameters, while OpenAI's o1-mini has 100 billion parameters. Alibaba also claimed that QwQ-32B excels in various benchmark platforms, including AIME 24, Live CodeBench, LiveBench, IFEval, and BFCL. Some tests indicated that QwQ-32B outperformed both DeepSeek R1 and OpenAI's o1-mini. For instance, in the LiveBench evaluation, which assesses AI model performance, QwQ-32B achieved a score of 73.1 percent, followed by DeepSeek R1 at 71.6 percent and o1-mini at 59.1 percent. 

---

Tag: Alibaba DeepSeek OpenAI