Singapore's national AI program has switched to Alibaba's Qianwan.
Hong Kong CTSC News, November 26th (Reporter Wang Fengling) - The Singapore National Artificial Intelligence Program (AISG) confirmed on the 25th that its ongoing Southeast Asian language large model project has completely abandoned the technical route based on the US Meta model and has instead adopted the open-source architecture of Alibaba's Tongyi Qwen.
According to the official news of AISG, the "Qwen-SEA-LION-v4" model launched by it on the 25th has quickly topped the authoritative open-source list in terms of measuring Southeast Asian language capabilities. It significantly outperforms previous models based on the US Meta Llama series in tasks such as semantic understanding and text generation for regional languages like Indonesian, Thai, and Malay. Moreover, the Qwen licensing terms are more conducive to local secondary development, thus the decision to switch the technical route was made.
AISG stated that Qwen has rapidly improved its performance since its full open-source in 2023. Currently, its download volume has exceeded 600 million times, surpassing models like Llama and Deepseek, and has become the most widely used open-source large model.
Professor Liu Chunsheng, an associate professor at the International Economics and Trade School of Central University of Finance and Economics, told Hong Kong CTSC on the 26th that Qwen has a good overall performance in terms of algorithms and computing power, and is expected to have absorbed a large amount of Southeast Asian language materials, having higher adaptability and a price advantage. It is highly attractive to Southeast Asian countries. Moreover, Qwen is deeply positioned in Southeast Asian e-commerce and other fields based on Alibaba's large ecosystem.
Since DeepSeek emerged this year, Chinese AI large models have redefined the development direction of the industry with open-source, low cost, and high performance. Recently, well-known Silicon Valley investor Chamath Palihapitiya announced that Kimi will replace OpenAI as a productivity tool; multiple programming platforms have integrated Zhi Pu models; Airbnb CEO stated that Alibaba Qwen is faster and more outstanding than the US OpenAI model. The Singapore National AI Program "changing its heart" is the latest achievement of Chinese open-source models in the global market.
The global AI competition landscape is constantly reshuffling. The latest comprehensive ranking of the LMArena Large Model Arena on November 26th shows that four models from Chinese companies are among the top 20 global AI large models.
Liu Chunsheng pointed out that the breakthroughs of Chinese AI models in technology blockades, and their accelerated overseas expansion have achieved remarkable results. There are multiple factors behind this. Chinese AI models are mostly free, making it easy to roll out, facilitating the collection of more data, continuously optimizing the processing speed and accuracy of text data, and improving local adaptability. The Chinese government's policy support is significant, and the research and development of consumer and industrial application-level large models have been well-organized and implemented, with good implementation effects. Artificial intelligence can naturally take over "Internet +", giving rise to more AI giant enterprises.
Liu Chunsheng believes that global AI development is beginning to show its brilliance, and the competition between the US and China will become more intense. Who will occupy the market share in each region in the future will test the comprehensive strength of various models in data processing and analysis capabilities, localization plans, trustworthiness, and cost performance. However, China and the US have both competition and cooperation. The overseas expansion of Chinese AI models, with their open-source and free nature, can also help developing countries and emerging industrialized countries to develop.