Chinese tech company Alibaba has launched its latest AI reasoning model QwQ-32B, a direct rival to DeepSeek’s R1 and OpenAI’s models.
The launch of QwQ-32B, currently in early public testing. has already impacted Alibaba stock performance. Following the announcement the company’s Hong Kong-listed shares surged by 8.39%, hitting a 52-week high, though its New York-listed stock saw a slight decline.
One of the standout features of QwQ-32B is its efficiency. While DeepSeek-R1 operates with 671 billion parameters – 37 billion of which are activated during inference – Alibaba’s model gets by with just 32 billion parameters.
This design reduces the computational power required to run the model. DeepSeek-R1 demands around 1600GB of VRAM, whereas QwQ-32B functions on just 24GB, making it accessible to a broader range of users, even those with high-end gaming GPUs like the Nvidia RTX 4090 or upcoming RTX 5090.
Alibaba claims that despite using fewer resources, QwQ-32B can match the performance of DeepSeek and OpenAI’s o1-mini. If true, this efficiency could make AI adoption more practical for businesses and developers seeking lower-cost solutions.
In early testing, QwQ-32B has shown promising capabilities. It provides detailed responses to queries, often giving extensive context and reasoning – sometimes to an excessive degree. This feature is somewhat similar to OpenAI’s Deep Thinking mode.
Like many AI models, QwQ-32B comes with restrictions. It actively flags political discussions and other sensitive topics as inappropriate, a censorship approach similar to that seen in DeepSeek and OpenAI’s models. While workarounds may emerge – as has been the case with DeepSeek – current safeguards remain in place.
