Table of Contents
On March 5, Chinese tech giant Alibaba released its latest AI reasoning model, QwQ-32B, resulting in an 8% spike in the company’s Hong Kong-listed shares. While less capable than America’s leading AI systems, such as OpenAI’s o3 or Anthropic’s Claude 3.7 Sonnet, the model reportedly performs about as well as its Chinese competitor DeepSeek’s model, R1, while requiring considerably less computing power to develop and to run. Its creators say QwQ-32B embodies an “ancient philosophical spirit” by approaching problems with “genuine wonder and doubt.”
[time-brightcove not-tgx=”true”]
“It reflects the broader competitiveness of China’s frontier AI ecosystem,” says Scott Singer, a visiting scholar in the Technology and International Affairs Program at the Carnegie Endowment for International Peace. That ecosystem includes DeepSeek’s R1 and Tencent’s Hunyuan model, which Anthropic co-founder Jack Clark has said is by some measures “world-class.” That said, assessments of Alibaba’s latest model are preliminary, both due to the inherent challenge of measuring model capabilities, and because so far, the model has only been assessed by Alibaba itself. “The information environment is not very rich right now,” says Singer.
Another step on the path to AGI
Since the release of DeepSeek’s R1 model in January sent waves through the global stock market, China’s tech ecosystem has been in the spotlight—particularly as the U.S. increasingly sees itself as racing against China to create artificial general intelligence (AGI)—highly advanced AI systems capable of performing most cognitive work, from graphic design to machine-learning research. AGI is widely expected to confer a decisive military and strategic advantage to whichever company or government creates it first, as such a system may be capable of engaging in advanced cyberwarfare or creating novel weapons of mass destruction (though experts are highly skeptical humans will be able to retain control over such a system, regardless of who creates it).
“We are confident that combining stronger foundation models with reinforcement learning powered by scaled computational resources will propel us closer to achieving AGI,” wrote the team behind Alibaba’s latest model. The quest to create AGI permeates most leading AI labs. DeepSeek’s stated goal is to “unravel the mystery of AGI with curiosity.” OpenAI’s mission, meanwhile, is to “ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity.” Leading AI CEOs including Sam Altman, Dario Amodei, and Elon Musk all expect AGI-like systems to be built within President Trump’s current term.
Read More: How China Is Advancing in AI Despite U.S. Chip Restrictions
China’s turn
Alibaba’s latest AI release comes just two weeks after the company’s co-founder, Jack Ma, was pictured in the front row at a meeting between President Xi Jinping and the country’s preeminent business leaders. Since 2020, when Ma publicly criticized state regulators and state-owned banks for stifling innovation and operating with a “pawn shop mentality,” the Chinese billionaire has largely been absent from the public spotlight. In that time, the Chinese government cracked down on the tech industry, imposing stricter rules on how companies could use data and compete in the market, while also taking more control over key digital platforms.
Singer says that by 2022, it became clear that the bigger threat to the country was not the tech industry, but economic stagnation. “That economic stagnation story, and attempting to reverse it, has really shaped so much of policy over the last 18 months,” says Singer. China is moving quickly to adopt cutting-edge technology, with at least 13 city governments and 10 state-owned energy companies reportedly having already deployed DeepSeek models into their systems.
Technical innovation
Alibaba’s model represents a continuation of existing trends: in recent years, AI systems have consistently increased in performance while becoming cheaper to run. Non-profit research organization Epoch AI estimates that the amount of computing power used to train AI systems has been increasing by more than 4x each year, while, thanks to regular improvements in algorithm design, that computing power is being used three times more efficiently each year. Put differently, a system that required, for example, 10,000 advanced computer chips to train last year could be trained with only a third as many this year.
Despite efficiency improvements, Singer cautions that high-end computing chips remain crucial for advanced AI development—a reality that makes U.S. export controls on these chips a continuing challenge for Chinese AI companies like Alibaba and DeepSeek, whose CEO has cited access to chips, rather than money or talent, as their biggest bottleneck.
QwQ (pronounced like quill) is the latest to join a new generation of systems billed as “reasoning models,” which some consider to represent a new paradigm in AI. Previously, AI systems got better by scaling both the amount of computing power used to train them and the amount and quality of data on which they were trained. In this new paradigm, the emphasis is on taking a model that has already been trained—in this case, Qwen 2.5-32B—and scaling the amount of computing the system uses in responding to a given query. As the Qwen team writes, “when given time to ponder, to question, and to reflect, the model’s understanding of mathematics and programming blossoms like a flower opening to the sun.” This is consistent with trends observed with Western models, where techniques that allow them to “think” longer have yielded significant improvements in performance on complex analytic problems.
Alibaba’s QwQ has been released “open weight,” meaning the weights that constitute the model—accessible in the form of a computer file—can be downloaded and run locally, including on a high-end laptop. Interestingly, a preview of the model, released last November, attracted considerably less attention. Singer notes that “the stock market is generally reactive to model releases and not to the trajectory of the technology,” which is expected to continue to improve rapidly on both sides of the Pacific. “The Chinese ecosystem has a bunch of players in it, all of whom are putting out models that are very powerful and compelling, and it’s not clear who will emerge, when it’s all said and done, as having the best model,” he says.
