Alibaba Cloud’s Qwen2.5-Max comes in seventh on Chatbot Arena, beating out DeepSeek-V3 in ninth place, but trailing DeepSeek-R1 in third.
When you spot a great trading opportunity but don’t have the funds ready, use contra trading limits to buy stocks or ETFs immediately without upfront cash. Trade now and settle later with the Tiger Cash Boost Account. Enjoy up to 7 days interest-free with a $20K credit limit. Plus, trade commission-free for up to 6 months, perfect for short-term trading. Plus, sign up and deposit to receive rewards up to $1000 and enhance your investment journey!
Chinese e-commerce giant Alibaba Group Holding’s latest open-source Qwen artificial intelligence (AI) model surpassed DeepSeek-V3 to become the top-ranked non-reasoning model from a Chinese developer, according to a third-party benchmarking and ranking platform, highlighting the rapid pace of Chinese firms’ advance in the emerging technology.
Alibaba Cloud’s Qwen2.5-Max, launched during the Lunar New Year holiday, has climbed to seventh place on Chatbot Arena, a benchmarking project developed by computer scientists from UC Berkeley to evaluate the performance of large language models (LLMs) – the technology underpinning chatbots such as OpenAI’s ChatGPT. Rankings are determined by users voting on the quality of the output. Alibaba owns the South China Morning Post.
DeepSeek-V3, which the namesake Chinese start-up launched in late December to the shock of the global tech community, is currently ranked ninth.
“Alibaba’s Qwen [2.5] Max is strong across domains,” Chatbot Arena said in a post on X. The model performs especially well in the technical areas, including coding, maths and hard prompts, which are used to elicit more direct and well-defined responses from chatbots, the organisation added.
It also said that the latest Qwen model matches top proprietary models such as Microsoft-backed OpenAI’s GPT-4o and Claude from Anthropic, backed by Amazon.com.
Qwen2.5-Max’s ascent reflects the intense competition among Chinese tech companies to advance AI capabilities and narrow the gap with leading US players. DeepSeek’s V3 and R1, a reasoning model released last month, have sent shock waves through the AI industry, hitting semiconductor stocks, due to their competitive performance at lower training and usage costs.
As interest in DeepSeek surged around the world, Alibaba quickly rolled out an update to Qwen on January 29, the first day of the Lunar New Year, when much of China was shut down for the week-long family holiday.
The latest Qwen2.5 model was a result of Alibaba Cloud’s exploration in the area of mixture of models (MoE), an innovation in LLM architecture to balance model efficiency and performance.
It was trained on more than 20 trillion tokens. In comparison, DeepSeek-V3 was trained on 14.8 trillion tokens. A token is the most basic unit of text or other input that is processed by generative AI models. Those trained on larger sets of tokens tend to perform better.
Still, Alibaba’s Qwen2.5 was ranked lower than DeepSeek-R1, which came in third, making it the highest-ranked Chinese AI model. Domestic and global firms have been rushing to adopt R1 on their platforms. Zhipu AI, another Chinese start-up based in Beijing, also secured a spot in the top 10.
Comments