The DeepSeek-V3.2-Exp model has made its official debut and is now available as open - source software. This innovative model incorporates a sparse Attention architecture. This design ingeniously cuts down on computational resource usage and boosts inference efficiency. Moreover, it leads to a substantial reduction in API pricing.

Following this release, three prominent companies—Huawei Ascend, Cambrian, and Hygon Information—concurrently declared that they had successfully adapted the model for their respective systems. Ascend accomplished the deployment leveraging its inference framework and generously open - sourced the relevant code. Cambrian managed to slash both training and inference costs through the implementation of a novel mechanism. Meanwhile, Hygon DCU achieved flawless adaptation and carried out in - depth optimization.

On the morning of September 30, the market experienced fluctuations but ultimately rallied. AI chips and stocks related to the Huawei Ascend concept showed robust performance. Industry analysts emphasized that this seamless integration of software and hardware signifies a pivotal shift in China's domestic AI ecosystem. It has transitioned from being merely "usable" to truly "user - friendly," creating a fully - fledged closed loop that spans from underlying computing power to upper - layer applications.

Guotai Haitong Securities is of the opinion that the widespread adoption of large models and generative AI in consumer terminals, along with the development of an AI ecosystem that harmoniously blends software and hardware, will serve as crucial benchmarks for assessing the growth potential of technology companies.