What is DeepSeek?

DeepSeek is a Chinese company that designs and trains language models. In 2025, it was still considered a startup by most analysts. Founded in 2023 in Hangzhou, China, it didn’t gain international recognition until January 2025.

The reasons for their rise to fame were the release of two language models that matched, and in some cases surpassed, the models of leading Western AI companies. They were not only more capable, but they also achieved the same results at a much lower cost. This last point is undoubtedly the most relevant.

However, DeepSeek is also the name of the chatbot application that uses the company’s language models to function. In fact, this term is almost always used to refer to the Android and iOS app or the platform’s website from which it’s possible to interact with DeepSeek’s powerful linguistic models.

The company’s latest models are these:

  • DeepSeek-V3 is a Language Model (L) that generates text from existing text. It responds immediately after a user submits a request, composing text according to the requested information. It is comparable to similar models from Western companies, such as Llama3.1, Claude-3.5, and GPT-4o. In fact, in some metrics, it surpasses them.
  • DeepSeek-R1 is a reasoning model. Although it also generates text from text, it was designed to simulate human reasoning, solving user-posed problems step by step. The idea is that the more computing time spent resolving the user’s request, the more reliable the answer will be.

One of the standout features of these models is their affordability. In fact, according to figures provided in the DeepSeek API documentation, using the most capable model, DeepSeek-R1, is approximately 50 times cheaper than using OpenAI’s o1 model. Furthermore, both are open source, so they can be modified and adapted to the needs of any company or organization.

What is DeepSeek used for?

As we’ve explained, when we use the term DeepSeek, we can refer to three things that, while related, aren’t the same. First, there’s the company DeepSeek, based in China and responsible for several language models. Currently, this company focuses primarily on AI research, although it offers a paid API for accessing its services.

As for its counterpart chatbot, it serves the same purpose as ChatGPT, interacting with language and reasoning models. Here are some DeepChatGPTs and tricks to help you get the most out of it. It’s important to understand that a chatbot, as such, is simply an interface that facilitates the use of the underlying technology —the AI ​​models. Think of the DeepSeek chatbot as the body of a car, into which various engines can be installed.

And these DeepSeek engines are currently DeepSeek-V3, an LLM without reasoning capabilities, and DeepSeek-R1, an LLM with reasoning capabilities. Both are open source and can be run locally using tools like Ollama. What are these models used for? If we focus on the first one, and without getting into technical details, we can say that it’s used to generate text from text. By simulating human neural networks, it determines which sets of words are most appropriate based on an input, the famous prompt.

For example, if the user tells the LLM that they have a cup and asks what they can fill it with, the model will respond that they can fill it with milk, coffee, or juice. This is the most statistically likely answer, even though a cup could technically also be filled with gasoline or oil.

To put it simply, the reasoning model forces the LLM to dedicate more computing time to resolving the user’s request. This allows for more accurate answers to complex problems, such as riddles, puzzles, or enigmas. It also helps solve mathematical problems with greater precision.

In our guides section,n you will find more practical information about DeepSeek. We detail how it works and how you can use it to generate images. We explain how to install it locally on your computer.