China’s DeepSeek Launches Preview of New AI Model

Last year, DeepSeek, a Chinese startup, made headlines with a budget-friendly chatbot that disrupted the AI landscape. Now, the company has unveiled its latest offering: the V4 model, which it claims showcases advanced capabilities. The preview version was shared on Friday, highlighting a major update with a context length of up to one million words.

DeepSeek-V4 comes in two iterations—Pro and Flash—each tailored for different needs. The Pro variant boasts 1.6 trillion parameters, while the Flash edition uses 284 billion, affecting the model’s ability to process and generate responses. The company emphasized that the Pro version outperforms other open-source models in global knowledge tests, nearly matching the performance of Google’s Gemini-Pro-3.1.

“In world knowledge benchmarks, DeepSeek-V4-Pro significantly leads other open-source models and is only slightly outperformed by the top-tier closed-source model, (Google’s) Gemini-Pro-3.1,” the company stated in a social media post.

By releasing early versions, DeepSeek aims to gather user insights before finalizing the model. This approach allows for adjustments based on real-world input. However, the startup has faced scrutiny over its chatbot’s handling of politically charged subjects. Reports indicate the AI tends to avoid detailed answers on topics like the 1989 Tiananmen crackdown, sparking debates about censorship.

Meanwhile, the United States has accused DeepSeek of engaging in “industrial-scale distillation campaigns to steal American AI,” a claim made by U.S. officials and rivals. Beijing has dismissed these allegations as unfounded, reaffirming its commitment to safeguarding intellectual property rights. The startup, based in Hangzhou, continues to navigate both praise and criticism as it advances its AI developments.

To view this video, please enable JavaScript and upgrade to a web browser supporting HTML5 video.