(Yicai) March 6 -- Shares in Alibaba Group Holding surged today after the Chinese internet giant released its latest open-source Tongyi Qianwen inference model, claiming that its overall performance is on a par with artificial intelligence startup DeepSeek’s large language model Deepseek-R1, which recently caused a global sensation.

Alibaba’s share price [HKG:9988] closed up 8.3 percent to HKD140.80 (USD18.10).

Tongyi Qianwen QwQ-32B has made significant advancements through large-scale reinforcement learning, achieving a big leap in its mathematics, coding and general capabilities, the Hangzhou-based firm said. Deployment costs have also been greatly reduced, making it possible to run on consumer-grade graphics cards for local deployment.

The QwQ-32B chatbot has 32 billion parameters, Alibaba’s Tongyi team said. It performed very well in a series of authoritative benchmark tests, nearly surpassing US AI firm OpenAI’s o1-mini model, which was released in September last year and is of a similar size. It can now rival DeepSeek-R1, which has 671 billion parameters, of which 37 billion are activated.

In addition, QwQ-32B integrates smart agent-related capabilities, enabling it to think critically while using tools and to adjust its reasoning process based on environmental feedback.

QwQ-32B is available for free downloads and commercialization, or users can experience it through the web-based Qwen Chat. The chatbot will also be available on the Tongyi app for free.

QwQ-32B not only offers strong reasoning power, it also requires less resources, thanks to reinforcement learning, the Tongyi team said. The team hopes to show that combining powerful foundational models with large-scale reinforcement learning could be a viable path toward achieving artificial general intelligence.

Editor: Kim Taylor