DeepSeek Usage Plunge? The Truth Behind the Headline Data
Sensational claims of “DeepSeek’s popularity collapse” are flooding Chinese social platforms, with videos alleging its usage nosedived from 50% to 3% amassing millions of views. Yet a closer look reveals a textbook case of data misinterpretation amplified by viral cycles.
Contrary to viral narratives, DeepSeek’s usage likely never approached 50%. Data from AI platform Poe shows Even at its January peak, DeepSeek captured approximately 6-7% of usage.
A current combined usage rate of around 3% for its R1 and V3 models – representing a roughly 50% decline from earlier highs – appears to have been misreported as a fall from 50%. This critical decimal point error ignited the “implosion” narrative.
The reported 50% drop warrants deeper context. AI usage metrics are inherently fragmented. While Poe shows 3%, rival platform OpenRouter frequently reports DeepSeek adoption exceeding 10%.
Critically, the centralized dip masks a tectonic shift in deployment strategy. DeepSeek’s aggressive open-source policy has led to widespread embedding by third parties – from Tencent Yuanbao and Baidu Search to smartphone assistants and smart home devices.
While web traffic declined, industry analysis reveals an explosive 20x growth in overall DeepSeek usage through third-party channels over six months. Users aren’t abandoning DeepSeek – they’re accessing it differently.
This decentralization appears intentional. DeepSeek’s native API, though aggressively priced, underperforms third-party implementations in latency and context length – areas the company seems to deprioritize.
Facing GPU constraints, DeepSeek CEO Liang Wenfeng has stated the company prioritizes AGI research over optimizing its commercial API. “We are not aiming to provide the best cloud service… but to achieve AGI,” he noted.
The narrative shift from collapse to strategic repositioning underscores a critical dynamic. For AI players like DeepSeek, current market share paints an incomplete picture. Their true competition lies in the marathon toward next-generation models – making the development timeline and capabilities of the anticipated DeepSeek-R2 a far more significant metric than cherry-picked usage stats.
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/4530.html
