Definition:
DeepSeek is a Chinese startup based in Hangzhou, creator of the Artificial Intelligence models DeepSeek-V3 and DeepSeek-R1. Unlike other models such as OpenAI, DeepSeek models are Open Source, which allows users with advanced knowledge to download their code from Github, enabling them to modify and improve the code. DeepSeek is developed through pure reinforcement learning and trial and error. It features a chatbot to which all kinds of questions can be asked, both from its own knowledge and from content that we can provide through documents.
Origin and evolution
The DeepSeek model, born in Hangzhou by the startup of the same name, and participated by shareholder Liang Wengeng, had its definitive breakthrough on January 27, 2025, causing significant value losses to tech companies like Nvidia and AMD, due to the disruptive nature of its cost structure and results. The DeepSeek assistant, based on DeepSeek-V3, arrived on the App Store on January 11, 2025, and in 3 weeks surpassed OpenAI’s ChatGPT in downloads. Its DeepSeek-V3 and DeepSeek-R1 proved to be on par with other more established ones like Meta and OpenAI. Its rapid growth has caused a stir in tech stock prices, as it meant generating similar results with a much lower investment, as well as China’s breakthrough in the race for Artificial Intelligence, until then led by the United States.
Characteristics of DeepSeek
The initial R1 model of DeepSeek has achieved superior results in performance and reasoning tests compared to the main open-source models, and very similar to models like OpenAI’s O1. It is a generative AI assistant trained with more than 600 million parameters. Its chatbot answers questions in a conversational format and solves complex problems through reasoning.
Advantages of DeepSeek
Among the advantages of DeepSeek we can cite:
- It is open source, which allows users to edit and customize its code, and presumably, evolve more quickly.
- It can be installed on any device, even on a local machine, so interactions are not recorded on the central server.
- Native, free chatbot with no subscription required.
- High reasoning capacity. The model offers great performance in mathematical and programming tests.
- Low development cost. It is estimated that the R1 model has cost between 90% and 95% less than its main competitors. This has shown that results can be achieved without the need for high-cost infrastructures.
Disadvantages of DeepSeek
We can highlight, however, some drawbacks of DeepSeek:
- Prior censorship in the model’s training: it does not respond to sensitive questions about the Chinese regime and its geopolitics.
- Uncertainty regarding the ownership and transparency of the model.