There is a new AI chatbot in town, garnering the interest of several parties, from casual enthusiasts, to the generative AI industry at large, to the finance space. 

The app-based chatbot is called DeepSeek. It was the No. 1 most downloaded free app on the iOS App Store in the U.S. over the weekend, beating out ChatGPT, which typically holds that spot, according to mobile app analytics firm Appfigures. Similarly, statistics indicated that mobile app downloads between the iOS App Store and Google Play Store were at approximately 2.6 million as of Monday.

Developed by a Chinese AI startup, DeepSeek has quickly garnered global attention for being an open-source generative-AI option among a sea of for-profit based services. News of the app’s fascinating introduction is swiftly pushing former AI darling’s storied beginnings to the background. Simply, the app offers performance that rivals known AI tools such as ChatGPT and Google Gemini for a fraction of the cost. It has also topped app download lists world-wide. 

In the grand scheme, its parent company, High-Flyer, established DeepSeek as part of its experimental research unit in April 2023– quickly breaking off into its own entity and standing by as an investor. At that same time the company released its DeepSeek-V2 model, which has been compared to products by several top Chinese AI companies, including ByteDance, Tencent, and Baidu. 

DeepSeek announced its V3 model in December 2024, detailing that it cost $5.6 million to train the model on Nvidia H800 chips, which aren’t as powerful as AI chips used in the U.S. In comparison, companies including OpenAI and Meta reported their billion-dollar investments into the leading chips from Nvidia. DeekSeek’s V3 model is said to be on par with Meta’s Llama 3.1 and OpenAI’s 4o models in terms of performance. 

The company introduced its latest technologies, first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1 last week, which has garnered DeepSeek and related entities considerable attention. In response, several stocks saw considerable losses on Monday, including Nvidia with an almost $600b drop in market value. Other stocks that suffered included Broadcom, Microsoft, and Alphabet.

The surge of new users to DeepSeek also caused server outages, and for the company to halt new registrations due to “large-scale malicious attacks.”
With a reasoning model in place, the chatbot is more on par with OpenAI’s most expensive Pro subscription, which costs an eye watering $200 per month. However, as an open-source model, developers can access the API for free, in addition to casual users tinkering with the app for free. Despite not revealing the dataset it used to train its R1 model, Fortune Magazine noted that DeepSeek likely used “reverse engineering” of other popular models to train its chatbot. This practice is common among many open source developers. Still, DeepSeek said the new models have high reasoning skills but struggle with readability and mixed language.