Remember DeepSeek? This Chinese startup previously made waves with its cost-efficient R1 model. Now, after a period of calm, it has released a long-awaited update for its AI model: DeepSeek V4. This new version, initially available as a preview, isn’t just about better performance. For the first time, DeepSeek is highlighting its transition away from Nvidia hardware. The company ran its training processes on Huawei’s Ascend chips to push for greater technological autonomy.
DeepSeek V4 AI challenges GPT-5.5 and Gemini 3.1-Pro with open-source approach
Many U.S. giants keep their most powerful models behind closed doors. However, DeepSeek keeps betting on an open-source approach. This allows developers worldwide to modify and build upon the core technology. This is the same strategy that helped the company gain massive traction from the beginning. According to data from OpenRouter, Chinese models like those from DeepSeek and Alibaba now account for nearly a third of global AI usage.
The real buzz surrounding DeepSeek V4 focuses on its “agent” capabilities. Unlike standard chatbots that simply answer questions, these AI agents can navigate complex workflows. DeepSeek claims the Pro version can outperform several high-end Western models in autonomous managing tasks, such as writing code and handling multi-step software interactions. The company also improved the context window to one million tokens. This allows the model to process and recall vast amounts of information.
Its coding skills are particularly impressive. Tests from Vals AI suggest that DeepSeek V4 currently leads the open-source pack in generating computer code. This lets human programmers get rid of repetitive tasks and focus on more important aspects.
Under surveillance in the west
DeepSeek is not above suspicion in the West. Big names, including OpenAI and Anthropic, have accused the startup of “distillation.” This is, imitate the behavior of other models by analyzing its outputs. Plus, several governments have restricted the use of the platform due to data privacy concerns.
Some analysts suggest that independent evaluations are still needed to confirm all of DeepSeek’s performance claims. Still, the pace of development of the firm’s releases shows that the competition is far from slowing down.
The post DeepSeek V4 Arrives: The Powerful, Open-Source Chinese AI Model That Doesn’t Need Nvidia appeared first on Android Headlines.
