Chinese company DeepSeek on Friday unveiled a preview of its long-awaited new artificial intelligence model, V4. Thanks to a new architecture that processes information more efficiently, it can handle much longer prompts and larger volumes of text than the company's previous-generation models.
Although V4 is an open-source model, its performance is comparable to that of leading proprietary models from Anthropic, OpenAI and Google. It is also DeepSeek's first model specifically optimized for Huawei's Ascend chips. This marks an important test of China's efforts to reduce its dependence on Nvidia technology.
One of V4's key features is its ability to work with longer text prompts, allowing the model to process larger amounts of information. With this, DeepSeek is seeking to compete not only on model performance but also on the technological infrastructure required to run them.
The significance of DeepSeek's new model goes beyond its performance benchmarks. The company is attempting to address two challenges simultaneously: developing artificial intelligence that can compete with leading systems and running it on Chinese-made chips. This highlights how competition in artificial intelligence is increasingly tied not only to software capabilities but also to access to computing resources.
The use of Huawei's Ascend chips is particularly noteworthy, as it provides an opportunity to assess how far Chinese companies can go in developing such complex systems using their own technological infrastructure. However, the launch of a single model does not mean that dependence on Nvidia has been eliminated. V4's true significance will become clear through its subsequent deployment and a comparison of its performance across different tasks.
