- DeepSeek's next-generation flagship V4 model will be released in late April, according to the Chinese AI company's founder.
- The highly anticipated new model is expected to achieve deep integration with domestic chips such as Huawei's Ascend for the first time.
DeepSeek's closely watched next-generation flagship V4 model is set to officially launch in late April, the Chinese AI startup's founder Liang Wenfeng recently said in an internal communication, according to a Friday report by local media outlet Chuangzhiji.
The highly anticipated new model will feature a trillion-parameter scale and a million-level context window, and will achieve deep integration with domestic chips such as Huawei's Ascend for the first time, the report said.
This marks a significant step forward for China’s AI industry as it moves away from reliance on traditional ecosystems and works to reduce its dependence on CUDA.
Currently, the training of mainstream AI models in China still relies heavily on Nvidia chips and requires iteration within the CUDA ecosystem. CUDA has become the de facto standard in the AI field, with a vast amount of algorithms and research built upon it.
In preparation for the V4 release, domestic tech giants including Alibaba, ByteDance, and Tencent have actively begun laying out underlying computing infrastructure, the report noted.
These companies have ordered hundreds of thousands of next-generation AI computing chips, planning to offer the new model via cloud services and deeply integrate it into their own AI products.
Driven by this demand, prices for next-generation AI chips have recently surged by about 20%, reflecting strong market expectations for the commercialization of large models and the domestic computing ecosystem.
Ahead of the official V4 launch, DeepSeek's web interface rolled out two interaction modes, "Quick Mode" and "Expert Mode," on April 8, marking the platform's first introduction of a tiered design since its rise to prominence.
Just in late March, the platform experienced a nearly 12-hour system crash, setting a record for its longest service outage and highlighting the computing power bottlenecks challenging the broader industry.
Upon the launch of the "Expert Mode," there was widespread industry speculation that the backend was likely utilizing some form of a lightweight test version of the upcoming V4 model.
As of now, DeepSeek has not publicly responded to the news of the V4 release, and the industry is awaiting its official announcement and the disclosure of technical details.
