DeepSeek Introduces V4 Model Optimized for Huawei Chips

DeepSeek presents an AI model for Huawei chip infrastructure (photo: dok. unsplash)

DeepSeek shared a preview of its latest Artificial Intelligence (AI) model, V4. This AI model is specifically designed to run on Huawei-made chip infrastructure.

Through this collaboration, the V4 model will operate at its maximum capacity on all Huawei system lines, including the Ascend supernode series. Previously, DeepSeek relied more on Nvidia chips, but now they are starting to switch to domestic solutions.

"All Ascend supernode product lines now support the DeepSeek V4 series model," Huawei said in a statement, quoted by Reuters on Friday, April 24.

In the world knowledge benchmark test, the pro version of this V4 model is claimed to be able to outperform various other open source models. DeepSeek V4 performance is reported to be only slightly behind Google's closed Gemini-Pro-3.1 model.

This launch was also accompanied by the release of a 'flash' version that offers a lower price for developers. Through this preview version, DeepSeek wants to collect user feedback before launching its final product in the near future.

This new model comes at a time when the White House has accused China of intellectual property theft of AI laboratories. DeepSeek has also been dragged into controversy over how they obtain chips and the source of data for training their models.

DeepSeek has denied allegations of illegal use of synthetic data and insists that their data is naturally collected through web searches. The Chinese Embassy in Washington also strongly rejected the allegations and expressed their commitment to protecting intellectual property rights.