Alibaba's AI model leads global crypto trading test, outperforms GPT-5

Alibaba's Qwen3-Max AI model earned top spot in a real-market crypto trading test, while DeepSeek AI came second with nearly 5 per cent gains

Don't want to miss the best from Business Standard?

Alibaba Group’s artificial intelligence model, Qwen3-Max, has outperformed five other major AI systems in a real-market cryptocurrency trading experiment conducted by US research firm Nof1, South China Morning Post reported. 

In the first round of the Alpha Arena competition which concluded on Tuesday, Alibaba Cloud’s Qwen3-Max achieved a 22.32 per cent return on a starting investment of $10,000 in just two weeks. Only two of the six competing models made profits.

  China’s DeepSeek V3.1 Chat ranked second with a 4.89 per cent gain, while all four US-developed models including OpenAI, Anthropic, Google DeepMind and Elon Musk’s xAI, recorded steep losses. OpenAI’s GPT-5 performed worst, losing 62.66 per cent of its investment, the news report said.

Experiment to test AI in ‘real-world’ conditions

Nof1 cautioned in a blog post that the early results “may be the result of luck”, noting that future rounds would include “more statistical rigour”. 

The Alpha Arena experiment gives AI models access only to quantitative market data, no news or external information, raising questions about how well such systems might perform in real-world investing. “If nothing else, we hope Alpha Arena highlights the power of evaluating AI in more consequential, realistic environments,” Nof1 said. “We believe this is the fastest path to uncovering critical gaps and insights that move frontier AI forward.”  Interestingly, Alibaba’s Qwen3-Max was the only model without reasoning capabilities, meaning it did not simulate step-by-step thinking before executing trades.

 

Alibaba steps up AI hardware development

Amid rising competition in artificial intelligence, Alibaba is also advancing its chip development efforts. According to an August report by The Wall Street Journal, the company has built a new AI chip that is said to be more versatile than previous versions.

  Unlike its earlier chip, which was manufactured by Taiwan Semiconductor Manufacturing Company (TSMC), the new model has been entirely produced in China and is still under testing. It is designed to handle a wider range of AI inference tasks, which involve running trained models efficiently.

 

China accelerates homegrown chip output

China’s major tech firms are focusing on domestic chip production as demand for AI computing surges. A Financial Times report noted that Chinese companies aim to triple the country’s AI chip output by 2026. 

The push comes amid rising tensions with the US over technology exports. Earlier this year, the Trump administration barred sales of Nvidia’s high-performance H20 chips in China, though the ban was lifted in July.  However, Chinese regulators later raised concerns about potential “information leaks” and backdoor access in Nvidia’s chips. Authorities reportedly advised major firms, including Tencent, ByteDance and Alibaba, to suspend H20 orders, prompting local companies to fill the gap.

  In response, Chinese players such as Huawei, Semiconductor Manufacturing International Corporation (SMIC), MetaX, Cambricon Technologies and Alibaba have accelerated efforts to expand chipmaking capacity. 

 

 

China still faces chipmaking challenges

Despite the rapid push, experts say China remains well behind US chipmakers. Industry insiders told the Wall Street Journal that the country’s factories still lack the capability to produce chips as advanced as those from the US, which face strict export restrictions from Washington. 

Even Nvidia’s H20, the most powerful chip allowed for sale in China, is significantly weaker than the H100 or Blackwell series, according to news agency Reuters. The absence of cutting-edge manufacturing technology continues to limit China’s ability to close the gap in high-performance AI hardware.