DeepSeek: what is it?

One of the technological surprises of 2025: DeepSeek's AI has revolutionised the sector with innovations over previous models. Find out more about this Chinese AI on our blog.

Communication Team

Telefónica Follow

Reading time: 3 min
  • DeepSeek’s strong emergence in early 2025 made it the most downloaded app in the App Store in the United States.
  • Its uses are varied, from AI chatbots to document processing, programming and financial analysis.
  • It differs from ChatGPT in terms of its open source code, philosophy and geographical origin.

DeepSeek is undoubtedly one of the technological stars of 2025, even becoming the most downloaded app in Apple’s App Store in the United States at the beginning of the year and also leading downloads in the Google Play Store in other markets.

This extremely high level of impact has been a turning point in the AI industry and has forced other players such as OpenAI and Nvidia to react to this new competitor.

What is DeepSeek

Founded in 2023, DeepSeek is a Chinese artificial intelligence company that, as we saw earlier, has achieved enormous popularity in its short lifetime.

It is based on open-source large-scale language models (LLMs), a topic that our colleagues at Telefónica Tech recently explored in depth in an interesting article called The incredible inner world of language models.

DeepSeek’s notoriety came both from its ability to compete head-to-head with global leaders in the field, such as ChatGPT, and from its greater efficiency in terms of costs and resources.

In addition to the use of AI chatbots, it also has other functionalities such as document processing, content generation and even more specialised areas such as programming and financial analysis.

Main features of DeepSeek

One of the distinguishing features of DeepSeek is that its models are open source, meaning that any company or individual can install and even modify them, allowing for control over data and the possibility of running them locally without relying on external servers.

In addition to the transparency of open source code, this AI model has been trained by pure reinforcement—that is, without labelled data—so its learning is autonomous through trial and error, which is then refined through supervised learning.

Although it is true that the initial boom was due to its English version, it is also available in other languages and, unlike other types of AI models, can be used without registration or subscription.

Finally, with regard to the models, the two used are DeepSeek V3 and DeepSeek R1, the latter based on the former but ‘enhanced through reinforcement learning’, which, as indicated on its website, offers the ability to solve mathematical and logical problems, scientific analysis and research projects.

Differences between DeepSeek and ChatGPT

As two of the most widely used AI models in the world, let’s look at some of the differences between them.

To begin with, and although it may seem obvious since we have already mentioned it, the geographical origin of the two is different: one is from China (DeepSeek) and the other from the United States (ChatGOT).

This difference goes beyond physical distance or the cultural peculiarities of each country, as it directly affects international relations and geopolitics, something that is potentially sensitive given the current global context.

On the other hand, there is also a difference in the original philosophy: while DeepSeek is based on open source and is free, ChatGPT operates under a closed, freemium model, offering basic features for free and reserving advanced options for paid users.

In terms of architecture, we could summarise the difference in that Asian AI uses a Mixture-of-Experts (MoE) approach, in which only the key parameters are activated for each task, while American AI uses a traditional transformer architecture, where all parameters are used for each function.

In this way, ChatGPT offers greater consistency, albeit with higher resource consumption, while DeepSeek reduces these costs and also increases efficiency.

With regard to privacy and security, there is another issue to consider. ChatGPT is governed by Western data protection standards, which may make it more attractive for regulated environments, this being one of the hypothetical weaknesses of Chinese AI.

Added to this is the detection of vulnerabilities such as unencrypted data transmission or database exposure, which has raised concerns about possible security flaws.

Share it on your social networks


Communication

Contact our communication department or requests additional material.