BEIJING: China technology companies are rushing to introduce advanced artificial intelligence (AI) reasoning models. These models are designed to “think and reflect” before giving responses, aiming to narrow the gap between China and the United States in AI technology.
According to the South China Morning Post, Alibaba Cloud, part of Alibaba Group Holding, recently showcased a preview of its reasoning model called QwQ.
Similar to OpenAI’s o1, which specialises in maths, programming, and scientific exploration, o1 was launched by the Microsoft-backed start-up in September.
Alibaba Cloud stated that QwQ matched or outperformed o1 in certain benchmark tests. The tech firm said it ranked higher in two maths evaluations and performed equally in problem-solving and coding.
The release of QwQ follows a series of reasoning model launches by other Chinese firms in recent weeks, which means Chinese companies have been catching up to their US counterparts.
Xu Liang, an AI entrepreneur from Hangzhou, said local firms are catching up with OpenAI while competing within China. He said that OpenAI’s o1 has “given the direction,” with more relevant research, Chinese companies will eventually improve in this area.
OpenAI’s o1 models, introduced in September, were designed to “spend more time thinking before they respond,” mimicking human reasoning. The o1 models refine their problem-solving process by trying different approaches and recognising their mistakes.
These allow them to handle complex tasks better than earlier generative AI models, according to OpenAI.
China companies have been quick to adapt. Tencent-backed Moonshot AI launched an updated version of its Kimi chatbot in early October. The new version called, “Explore,” incorporates reasoning capabilities and expanded search functions.
When asked to compare investments in gold and Chinese electric vehicle giant BYD, Kimi analysed stock and gold prices before delivering a calculated answer within minutes.
Other companies have also joined the race. Hangzhou-based DeepSeek introduced its r1 model, outperforming OpenAI’s o1 in three out of six benchmarks evaluating maths, programming, and scientific tasks.
Shanghai AI Lab also developed an enhanced process for its reasoning model InternThinker, focusing on understanding queries, recalling knowledge, planning tasks, self-reflection, and summarising.
Gaming company Kunlun Tech, which owns the Opera web browser, and Alibaba’s international commerce unit have also released reasoning models.
These launches occurred within the last two weeks, just days apart, highlighting the fast-paced competition among firms in China.
While these models represent significant progress, DeepLearning.AI, an AI education company, warned that longer response times make these models less practical for daily tasks, especially due to their higher costs.
The company noted on its website that reasoning models excelling in maths and science are often “slow and costly.” /TISG
Featured image by Depositphotos (for illustration purposes only)