DeepSeek has introduced a preview of its new AI model, V4, showing China’s growing effort to rely on its own technology instead of foreign systems.

The new model has been specially adapted to work with Huawei chips, especially the Ascend processors, which are seen as China’s strongest local alternative to Nvidia. This is a major shift, as DeepSeek previously depended heavily on Nvidia hardware.

According to the company, the Pro version of V4 performs better than most open-source AI models in knowledge-based tests, coming close to top closed models like Google’s Gemini-Pro-3.1.

Experts say this collaboration is an important step for China’s AI industry. It shows that advanced AI models can now be trained and run using local chips instead of imported ones.

The model has quickly gained attention and climbed to the top trending spot on Hugging Face, a popular platform for sharing AI models. It is especially strong in handling long and complex text tasks while keeping costs lower than competitors. However, it currently lacks features like image and video processing.

This development comes at a time when tensions between the US and China are affecting the tech industry. Restrictions on advanced chips have pushed Chinese companies to speed up their efforts to become self-reliant.

Despite its success, DeepSeek faces challenges. Some countries have raised concerns about data privacy and restricted its use. At the same time, competition inside China is increasing, with companies like Zhipu AI and MiniMax also launching strong alternatives.

The V4 model is also designed for advanced AI agents that can perform more complex tasks than regular chatbots, though these require more computing power.

While early results look promising, experts say the model still needs more real-world testing before its full impact can be judged.