According to the “Q2 2025 AI Application Value Ranking” released by QuestMobile， DeepSeek， which held the top spot in Q1， has seen a significant decline in its average monthly downloads. Downloads plummeted from a remarkable 81.113 million in Q1 to 22.589 million in Q2， a sharp decrease of 72.2%. This has led some netizens to comment that DeepSeek has “fallen from grace.”
However， industry insiders offer a different perspective. They argue that many highly engaged users rarely log in to official websites or apps. Instead， they access AI services through interfaces provided by various service providers， such as Tencent Yuanbao， Baidu， Huoshan， and WeChat mini-programs. Furthermore， some organizations have proactively integrated DeepSeek into their systems and deployed it locally. These usage scenarios， which represent a substantial portion of DeepSeek’s adoption， are not captured in the download statistics. Rather than a fall from grace， this shift can be seen as DeepSeek “entering ordinary households.”
Industry experts point out that DeepSeek has emerged as a disruptive force in the AI landscape， employing a strategy of “low cost， high performance， and full open-source.” This approach has redefined the “cost-effectiveness” of AI research and development， successfully tackling the “impossible triangle” of large model training – achieving high performance， low cost， and ease of use simultaneously.
The widespread global popularity of DeepSeek’s large models， particularly its open-source nature， can be attributed not only to its optimized algorithmic architecture and significant reduction in training costs but also to its departure from traditional paradigms in AI large model development. By making its models open-source， DeepSeek has provided developers with extensive opportunities for secondary development， thereby promoting the widespread adoption and democratization of artificial intelligence applications.
