DeepSeek's new V3-0324 model puts pressure on OpenAI
With this latest upgrade, DeepSeek is positioning itself even closer to its Western competitors.
At this point, it’s not even surprising when the hottest new AI model doesn’t come from Silicon Valley. A few years ago, that would’ve been unthinkable. Now? The world’s biggest AI breakthroughs could just as easily come out of Shenzhen or Hangzhou as San Francisco. That’s exactly what’s happening with DeepSeek.
This week, the Chinese AI startup rolled out a major upgrade to its large language model, and the buzz around it is almost as loud as when DeepSeek first made waves earlier this year.
The latest version, V3-0324, is now live on Hugging Face, and the company says it delivers “significant improvements” over its predecessor, V3—particularly in reasoning, coding, and Chinese-language tasks. It’s yet another sign that DeepSeek is pushing harder into OpenAI’s territory.
DeepSeek is moving fast. Launched just last year, it had already released its first big model, V3, in December. Then, in January, it dropped R1—a research-focused model that put China in direct competition with U.S. AI giants like OpenAI and Meta. R1 even outperformed OpenAI’s o1 model in certain benchmarks, including the MATH-500 test, and the open-source version shot to the top of the iPhone app charts, briefly surpassing ChatGPT.
Oyinebiladou Omemu
Now, with this latest upgrade, DeepSeek is positioning itself even closer to OpenAI’s GPT-4 or Anthropic’s Claude 2—but at a fraction of the cost.
DeepSeek says its latest model brings better performance in long-form writing, front-end web development, and interactive rewriting. It also boasts a nearly 20-point improvement on the AIME (American Invitational Mathematics Examination) benchmark. If DeepSeek’s models meet up with these claims and are significantly cheaper, a lot of businesses will start looking at their options, putting pressure on its Western competitors.
Of course, not everyone is thrilled about that. The U.S. government has already started throwing up roadblocks, recently banning DeepSeek’s chatbot from government-issued devices after linking it to a blacklisted Chinese state-run telecom company. There’s already speculation that a broader ban could follow. If that happens, it wouldn’t be the first time Washington has tried to shut out a rising Chinese tech company.
But DeepSeek isn’t some unknown startup hoping to make a splash. It’s already proving it can compete. And if OpenAI and its U.S. rivals don’t keep up, they might find themselves not only fighting China’s AI push—but also struggling to justify their own sky-high price tags.
Louis Eriakha
