Chinese technology company Alibaba Group Holding Ltd (阿里巴巴) yesterday released a new version of its Qwen 2.5 artificial intelligence (AI) model that it said surpassed the highly acclaimed DeepSeek-V3.
The unusual timing of the Qwen 2.5-Max’s release, on the first day of the Lunar New Year when most Chinese people are off work and with their families, points to the pressure Chinese AI start-up DeepSeek’s (深度求索) meteoric rise in the past three weeks has placed on not just overseas rivals, but also its domestic competition.
“Qwen 2.5-Max outperforms ... almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B,” Alibaba’s cloud unit said in an announcement posted on its official WeChat account, referring to OpenAI and Meta Platforms Inc’s most advanced open-source AI models.
Photo: Bloomberg
The Jan. 10 release of DeepSeek’s AI assistant, powered by the DeepSeek-V3 model, as well as the Jan. 20 release of its R1 model, has shocked Silicon Valley and caused tech shares to plunge, with the Chinese start-up’s purportedly low development and usage costs prompting investors to question huge spending plans by leading AI firms in the US.
However, DeepSeek’s success has also led to a scramble among its domestic competitors to upgrade their own AI models.
Two days after the release of DeepSeek-R1, TikTok owner ByteDance Ltd (字節跳動) released an update to its flagship AI model, which it said outperformed Microsoft Corp-backed OpenAI’s o1 in AIME, a benchmark test that measures how well AI models understand and respond to complex instructions.
This echoed DeepSeek’s claim that its R1 model rivaled OpenAI’s o1 on several performance benchmarks.
The predecessor of DeepSeek’s V3 model, DeepSeek-V2, triggered an AI model price war in China after it was released in May last year.
The fact that DeepSeek-V2 was open-source and unprecedentedly cheap, only 1 yuan (US$0.14) per 1 million tokens — or units of data processed by the AI model — led to Alibaba’s cloud unit announcing price cuts of up to 97 percent on a range of models.
Other Chinese tech companies followed suit, including Baidu Inc (百度), which released China’s first equivalent to ChatGPT in March 2023, and the country’s most valuable Internet company, Tencent Holdings Ltd (騰訊).
Liang Wenfeng (梁文鋒), DeepSeek’s enigmatic founder, said in a rare interview with Chinese media outlet Waves in July last year that the start-up “did not care” about price wars and that achieving AGI (artificial general intelligence) was its main goal.
OpenAI defines AGI as autonomous systems that surpass humans in most economically valuable tasks.
Liang said he believed China’s largest tech companies might not be well suited to the future of the AI industry, contrasting their high costs and top-down structures with DeepSeek’s lean operation and loose management style.
“Large foundational models require continued innovation, tech giants’ capabilities have their limits,” he said.
AI SERVER DEMAND: ‘Overall industry demand continues to outpace supply and we are expanding capacity to meet it,’ the company’s chief executive officer said Hon Hai Precision Industry Co (鴻海精密) yesterday reported that net profit last quarter rose 27 percent from the same quarter last year on the back of demand for cloud services and high-performance computing products. Net profit surged to NT$44.36 billion (US$1.48 billion) from NT$35.04 billion a year earlier. On a quarterly basis, net profit grew 5 percent from NT$42.1 billion. Earnings per share expanded to NT$3.19 from NT$2.53 a year earlier and NT$3.03 in the first quarter. However, a sharp appreciation of the New Taiwan dollar since early May has weighed on the company’s performance, Hon Hai chief financial officer David Huang (黃德才)
The Taiwan Automation Intelligence and Robot Show, which is to be held from Wednesday to Saturday at the Taipei Nangang Exhibition Center, would showcase the latest in artificial intelligence (AI)-driven robotics and automation technologies, the organizer said yesterday. The event would highlight applications in smart manufacturing, as well as information and communications technology, the Taiwan Automation Intelligence and Robotics Association said. More than 1,000 companies are to display innovations in semiconductors, electromechanics, industrial automation and intelligent manufacturing, it said in a news release. Visitors can explore automated guided vehicles, 3D machine vision systems and AI-powered applications at the show, along
FORECAST: The greater computing power needed for emerging AI applications has driven higher demand for advanced semiconductors worldwide, TSMC said The government-supported Industrial Technology Research Institute (ITRI) has raised its forecast for this year’s growth in the output value of Taiwan’s semiconductor industry to above 22 percent on strong global demand for artificial intelligence (AI) applications. In its latest IEK Current Quarterly Model report, the institute said the local semiconductor industry would have output of NT$6.5 trillion (US$216.6 billion) this year, up 22.2 percent from a year earlier, an upward revision from a 19.1 percent increase estimate made in May. The strong showing of the local semiconductor industry largely reflected the stronger-than-expected performance of the integrated circuit (IC) manufacturing segment,
NVIDIA FACTOR: Shipments of AI servers powered by GB300 chips would undergo pilot runs this quarter, with small shipments possibly starting next quarter, it said Quanta Computer Inc (廣達), which supplies artificial intelligence (AI) servers powered by Nvidia Corp chips, yesterday said that AI servers are on track to account for 70 percent of its total server revenue this year, thanks to improved yield rates and a better learning curve for Nvidia’s GB300 chip-based servers. AI servers accounted for more than 60 percent of its total server revenue in the first half of this year, Quanta chief financial officer Elton Yang (楊俊烈) told an online conference. The company’s latest production learning curve of the AI servers powered by Nvidia’s GB200 chips has improved after overcoming key component