The latest IT news still revolves around artificial intelligence. Companies are competing to create the most powerful AI models. American products like OpenAI's ChatGPT and Google's Gemini are already well-known worldwide. But in early 2025, Chinese AI developments made a big splash in the news.
While Donald Trump announced a $500 billion investment in AI, China responded with significant funding. The country plans to invest 1 trillion yuan (about $137.4 billion) in developing its own AI models.
What are these new Chinese AI models? Should users consider switching from American AI to Chinese alternatives? Our article will explore all these questions.
DeepSeek V3: A ChatGPT Killer?
The launch of the Chinese AI model DeepSeek V3 has shaken the tech world. No one expected such a significant breakthrough, especially from a small Chinese company founded by Liang Wenfeng in December 2023.
The company claims that its AI can do everything OpenAI's GPT-4o does but at a much lower cost. Training DeepSeek V3 reportedly cost around $6 million, while OpenAI spent over $100 million on ChatGPT. Building this advanced AI was cheaper than a top manager's annual salary in major U.S. tech companies.
How did Chinese engineers reduce costs so much? DeepSeek uses 2,000 high-performance chips, whereas American models use 16,000 high-end chips. Developers also enabled self-learning capabilities, which sped up the training process and reduced expenses.
On the day of its release, DeepSeek V3 became the #1 app in the U.S. App Store. At the same time, the stock market plunged into chaos. Nvidia's shares dropped by 17%, causing the company to lose nearly $400 billion—the largest single-day loss in stock market history.
This collapse triggered a domino effect. Investors panicked and started selling off shares of other major tech companies, including Microsoft, Google, Oracle, and Netflix.
However, the panic began to fade as analysts examined DeepSeek V3 more closely. The next day, Nvidia's stock rebounded by 6%, and Microsoft's shares rose above their pre-crash level. Experts agreed that the hype around DeepSeek was overblown and that initial fears were exaggerated.
OpenAI Strikes Back
Just two days later, on January 29, OpenAI responded. Teaming up with Microsoft, the company accused DeepSeek's developers of using distillation—a method where a smaller AI model is trained using data from a more powerful one. OpenAI claimed that DeepSeek had copied its technology, though this approach is common in AI development.
There were also doubts about DeepSeek's surprisingly low training costs. Some experts suggested that the model may have been trained on ChatGPT's existing data, which would have significantly reduced expenses.
Meanwhile, AI enthusiasts rushed to compare DeepSeek with ChatGPT. Their conclusion? The two models performed almost identically.
Are There Any Downsides to DeepSeek?
One primary concern is that DeepSeek follows strict Chinese censorship rules. For example, when a BBC journalist asked about Tiananmen Square, the model refused to answer.
Privacy is another issue. Experts worry about how DeepSeek handles user data.
According to its privacy policy, DeepSeek collects a large amount of personal information stored on Chinese servers. This data may include:
- Your email, phone number, and date of birth (from your account details)
- Anything you type or say, including chat history and voice inputs
- Technical information, such as your device model, operating system, IP address, and even your typing patterns
DeepSeek also shares this data with service providers and advertisers without setting a time limit for how long it will be kept.
Experts’ concerns about DeepSeek’s security were confirmed quickly. On January 30, Wiz Research discovered a database containing sensitive information. It included user chat history, API keys, logs, and backend data with internal details about the infrastructure.
This vulnerability allowed attackers to take complete control of the system without authentication.
However, OpenAI is also not perfect when it comes to privacy. Emily Taylor, CEO of Oxford Information Labs, advises users to be careful when sharing sensitive information with any AI model, not just DeepSeek.
DeepSeek isn't the only Chinese AI model challenging American companies. Alibaba, a major player in AI, has also released a new model—and some experts say it's even more powerful than DeepSeek.
Qwen2.5-Max: The New AI Leader
On Chinese New Year, Alibaba released Qwen2.5-Max, a powerful AI model that has outperformed DeepSeek V3 and GPT-4o.
This new model was trained on over 20 trillion tokens and uses advanced post-training techniques. For comparison, the Pro version of ChatGPT, which costs $200 per month, has a token limit of just 128,000.
Qwen2.5-Max was designed for efficient scaling, focusing strongly on logic and reasoning.
What can Qwen2.5-Max do?
- Answers university-level exam questions
- Writes high-quality code
- Creates games
- Finds real-time information
- Analyzes images and videos
And that's just a small part of its abilities. Alibaba claims it is significantly more advanced than DeepSeek V3.
“Qwen2.5-Max outperforms DeepSeek V3 in benchmarks such as Arena-Hard, LiveBench, LiveCodeBench, and GPQA-Diamond, while also demonstrating competitive results in other assessments, including MMLU-Pro,” – claims Alibaba.
However, the model is not open-source, meaning the public cannot study its inner workings. Access is only available through Alibaba's cloud service or Qwen Chat.
Right now, experts consider Qwen2.5-Max the most powerful AI model. But as AI technology evolves rapidly, breakthroughs won't be far behind.
Conclusion
Competition in the AI market is driving real progress. As countries race to lead in technology, we benefit from powerful AI tools that simplify life and business.
New AI models create exciting opportunities for personal use and business growth. If you want to integrate AI into your project and make the most of it, Vilmate is here to help.
Our AI experts will turn your ideas into reality and give you a decisive advantage over competitors.
