In brief: Alibaba has struck back at rival DeepSeek with the surprise release of its new Qwen 2.5-Max model. The Chinese e-commerce titan claims its latest artificial intelligence offering surpasses the capabilities of DeepSeek's recently launched and highly-touted DeepSeek-V3.
The timing of the Qwen 2.5-Max's debut is unusual, considering it arrived on the first day of the Lunar New Year holiday, when most Chinese workers are off. It illustrates just how severely DeepSeek's AI breakthrough has rattled the established players.
We've seen the effect DeepSeek's breakthrough had on overseas rivals like OpenAI, leading to multiple posts on X by CEO Sam Altman and the massive $600 billion stock crash at Nvidia – the biggest single-day plunge for any public company ever. It's no surprise that DeepSeek's success also spurred powerful domestic Chinese tech giants to scramble for a response. Alibaba's counterpunch comes in the form of the new Qwen 2.5-Max.
"Qwen 2.5-Max outperforms...almost across the board GPT-4o, DeepSeek-V3 and LLaMa-3.1-405B," boasted Alibaba Cloud in its WeChat announcement, calling out some of the most advanced open-source AI models from the likes of OpenAI and Meta.
Beyond DeepSeek's general AI capabilities, another factor that contributed to its popularity has been the extremely low costs of developing and running its models. This has even led investors to seriously question the massive spending on AI by US tech leaders.
Likely taking that into account, Alibaba Cloud also emphasized Qwen 2.5-Max's efficiency in a blog post, highlighting that it was trained on over 20 trillion tokens while using a mixture-of-experts (MoE) architecture that requires significantly fewer computational resources than usual approaches.
Beyond Alibaba, TikTok parent ByteDance has responded with an updated version of its flagship AI, which it claims outperformed OpenAI's GPT-3.5 on certain benchmarks.
An earlier version of DeepSeek also triggered an intense price war in China back in May. DeepSeek-V2's incredibly low cost of just 1 yuan (14 cents) per million tokens of data processed forced major cloud providers like Alibaba to slash their own AI model pricing by up to 97%.
It's worth mentioning that, like DeepSeek, Alibaba's new Qwen 2.5-Max does seem to avoid discussing sensitive political topics related to China. Attempts to query it on such issues are reportedly met with messages about exceeding data quotas, even as it responds normally to other prompts.