Just as US AI innovators scramble hastily to counter the impact of Chinese AI wunderkind, DeepSeek, another Chinese company has issued a challenge.
Tech giant Alibaba has launched a new version of its Qwen2.5 AI LLM and is making the claim that it is faster and more efficient than DeepSeek-V3.
The Chinese company though also took aim at OpenAI and Meta in a move that will rile President Trump, stating that its new launch also “outperforms … almost across the board GPT-4o… and Llama-3.1-405B.”
What Is Qwen 2.5-Max?
This is the latest model to join the Qwen 2.5 family, which are built using both proprietary and open-source technology, in contrast to rivals who have opted for one or the other option. OpenAI and Baidu – another Chinese AI contender – have both largely used closed source approaches while DeepSeek’s agile and relatively small team uses an open source approach.
This just in! View
the top business tech deals for 2025 👨💻
The Qwen 2.5 family was originally launched in May but now has more than 100 models, which range in size from 0.5 to 72 billion parameters. “Our base models have demonstrated significant advantages across most benchmarks,” Alibaba stated, “and we are optimistic that advancements in post-training techniques will elevate the next version of Qwen 2.5-Max to new heights.” It adds that it delivers pretraining on over 20 trillion tokens.
Race to the Top
This latest launch arrived on the first day of the Lunar New Year, when much of China is on holiday, which has prompted speculation that its announcement was brought forward because of the hype surrounding DeepSeek since its launch on January 10.
Its arrival caused tech shares to plummet – notably in Nvidia and Microsoft – though BBC News is reporting that they have now stabilized.
But the launch also, says Reuters, resulted in barbed questions about why AI development is coming at such a huge cost for companies outside of China. DeepSeek’s team is reportedly small and their costs are minimal compared to companies like OpenAI and Meta, but their product is outperforming rivals.
Panic in China and Beyond
DeepSeek’s arrival has caused ripples in its domestic market – where it is competing with Baidu and Alibaba. Tellingly, TikTok owner, ByteDance, released an update to its own flagship AI model just days after DeepSeek V.3’s arrival. It came with claims that it could outperform OpenAI’s o1 model in a benchmark test that specifically measures how AI models understand and then respond to complex instructions.
Meanwhile, in the US, President Trump is pushing a fast and furious AI agenda with his $500Bn AI Infrastructure Project, Stargate. It has kicked off with the launch of the Super Agent ChatGPT Gov from OpenAI.
But this launch was peppered with language squarely aimed at the Chinese. OpenAI CPO Kevin Weil said all of OpenAI’s innovation – is underscored by “how important it is that the US wins this race.”
But the DeepSeek team is simply shrugging this off. The company’s founder, Liang Wenfeng, told Chinese media outlet Waves in July that the startup “did not care” about price wars and that its goal was simply achieving AGI (artificial general intelligence). He added: “Large foundational models require continued innovation, tech giants’ capabilities have their limits.”
The implication is that tech giants, wherever they are based, can’t compete with this agile and fast innovating company, however much Government backing and money they throw at development.
