10 questions about DeepSeek everyone asks

In just a few days, DeepSeek has become the most talked-about AI research company, shaking up the U.S. tech ecosystem in a way rarely seen. Today, we reveal the keys to this developer and its success.

1. What is DeepSeek
2. Who created DeepSeek
3. How much does DeepSeek cost
4. How to download DeepSeek
5. How to use DeepSeek
6. Can DeepSeek generate images?
7. Is it actually better than ChatGPT?
8. Is it more efficient than other AIs?
9. Does DeepSeek have censorship?
10. How did Deepseek affected other technologic companies?

What is DeepSeek?

DeepSeek is a Chinese startup specializing in artificial intelligence research and the development of open-source models. The products it develops include both language models and data processing systems. Its most recent models, which have brought it widespread recognition, are DeepSeek R1 and DeepSeek V3.

DeepSeek R1 is an open-source reasoning model designed to handle reasoning tasks and solve complex problems. This artificial intelligence has revolutionized the market, as its performance surpasses that of o1 by OpenAI in several parameters related to efficiency, mathematical problem-solving, and programming.

On the other hand, DeepSeek V3 is a language model optimized for processing natural language and performing text comprehension and generation tasks. It is based on a MoE (Mixture-of-Experts) architecture, which allows it to divide its learning into multiple “experts” specialized in specific tasks, thereby achieving superior overall performance. DeepSeek V3 has outperformed models such as Llama 3.1 405B, Claude 3.5, and GPT-4o in several parameters.

Who created DeepSeek?

DeepSeek was founded in 2023 by Liang Wenfeng, a businessman with a strong interest in technology. He was born in Guangdong, China, in 1985. Wenfeng began his career as an investor and, in 2008, pursued studies in computer vision at Zhejiang University, located in Hangzhou. Later, he combined his two interests by researching how to apply artificial intelligence to stock trading.

In 2015, he founded High-Flyer Capital Management, a hedge fund company where he implemented algorithms to evaluate investment opportunities. By 2017, almost all of his investments were made using artificial intelligence. Today, High-Flyer Capital Management is the sole company funding DeepSeek.

How much does DeepSeek cost?

Currently, DeepSeek offers several services with different capabilities. The main ones are as follows:

1. DeepSeek web chatbot: Based on V3 model technology, it is free to access. The only requirement is to register in order to start interacting with the artificial intelligence. Recently, the company introduced a feature called “DeepThink,” which uses the capabilities of R1 to provide complex responses through an advanced reasoning process.
2. DeepSeek app: This also utilizes the V3 artificial intelligence model, includes the “DeepThink” option, and functions as a conversational chat. It is completely free to use.
3. R1 model API:
   • Input cost: $0.55 USD (€0.53) per million tokens if the input data is not in the cache and needs to be processed for the first time.
   • Cached input cost: $0.14 USD (€0.13) per million tokens if the information has already been processed and is stored in the cache.
   • Output cost: $2.19 USD (€2.10) per million tokens generated.

How to download Deepseek

You can download the DeepSeek app for Android and iOS from the Google Play Store, the Apple App Store, or directly from the company’s website using a QR code. After installation, simply register with your email, Google account, or Apple ID.

If you already have a web version account, your profiles and history will automatically sync.

Introducing DeepSeek App!

Powered by the world-class DeepSeek-V3
FREE to use with seamless interaction
Now officially available on the App Store, Google Play, and major Android markets
Download now: https://t.co/DIwqqkbK93

1/3 pic.twitter.com/QorHk2lEwj

— DeepSeek (@deepseek_ai) January 15, 2025

For those interested in obtaining the DeepSeek R1 API, go to their website and register. Afterward, select the “API Platform” option located at the top right of the screen.

Once on the platform, click on “API Keys” in the left sidebar to create a new key, which you can use to integrate the API into your system.

How to use DeepSeek

Using the DeepSeek chatbot is straightforward and user-friendly, as it functions similarly to other conversational AI models like ChatGPT.

The chatbot can be accessed via the web or the app. In both cases, registration is required to start using this AI for free. Once registered, you will enter the main chatbot interface, which includes a conversation panel and a dropdown menu on the left side for starting new chats, accessing past conversations, downloading the app, and managing profile settings.

To interact with DeepSeek, just type a prompt in the text box. You can also click the clip-shaped button to attach text documents or images.

Activate the “Search” option beneath the text box to let the AI search the web for information. Next to this is the “DeepThink” function, which uses the DeepSeek R1 reasoning model for solving complex problems or queries.

In “DeepThink” mode, responses may take longer as the AI shows the steps and logic used to reach a solution. Additionally, it will display the exact time taken to complete its reasoning at the top of its response.

Can DeepSeek generate images?

Unfortunately, DeepSeek is not capable of generating images at this time, and its features are limited to conversational tasks. However, you can use the chatbot to create detailed and accurate descriptions to input into an AI image generator.

How DeepSeek R1 measures up against ChatGPT and other AI models

The developers of DeepSeek have shared results from tests on their DeepSeek R1 model, comparing them with some competitors. They’ve claimed that R1 outperforms OpenAI’s o1 model in several areas.

In particular, R1 has shown superior performance over o1 in parameters such as AIME (which measures efficiency), MATH-500 (for mathematical problems), and SWE-bench Verified (for programming tasks).

It’s important to note that o1 is among OpenAI’s most advanced models, featuring sophisticated reasoning abilities, and is available in certain ChatGPT paid plans. In contrast, R1’s capabilities are included in DeepSeek’s free chatbot accessible via their website, app, and API.

However, ChatGPT does offer image generation and other advanced features, such as creating custom and specialized GPTs or accessing them through the GPT Store. Despite DeepSeek R1’s ability to outperform one of ChatGPT’s powerful integrated models, both are excellent options.

Is it more efficient than other AI models?

DeepSeek has gone beyond just comparing itself with OpenAI, evaluating its performance against other leading AI products. For its language model, DeepSeek V3, tests indicate that it surpasses models such as Llama 3.1 405B, Claude 3.5, or GPT-4o in various parameters.

Introducing DeepSeek-V3!

Biggest leap forward yet: ⚡ 60 tokens/second (3x faster than V2!) Enhanced capabilities API compatibility intact Fully open-source models & papers

1/n pic.twitter.com/p1dV9gJ2Sd

— DeepSeek (@deepseek_ai) December 26, 2024

Does DeepSeek Have Censorship?

DeepSeek is required to undergo evaluations by China’s Internet regulator, which means it must align with what they refer to as “fundamental socialist values.” As a result, this AI avoids or declines to answer certain questions that are deemed to be on sensitive topics.

However, it’s worth noting that this censorship exists only within DeepSeek’s chatbot. If you use its API and install the model on your computer, you won’t encounter these ideological restrictions.

Impact on Other Tech Companies

Until just a few days ago, the technological landscape in artificial intelligence was dominated by a few major players, with OpenAI and Google standing out the most. However, the emergence of a new competitor has shaken the industry to such an extent that all the major players have been affected.

DeepSeek has not only released open-source models that are as competitive as OpenAI’s technology, but these models also require lower-capacity chips to function. This factor has not gone unnoticed and has led to increasing scrutiny of billion-dollar investments in artificial intelligence projects and the necessity of using large quantities of chips for their development.

This has had a significant impact on Nvidia, a technology company that develops GPUs and chips to power artificial intelligence systems. Nvidia’s stock has dropped 17%, causing it to lose approximately $600 billion in market value. As a result, Nvidia has fallen from being the most valuable company in the world to third place, with Apple (1st place) and Microsoft (2nd place) surpassing it.

Additionally, the stocks of other major technology companies have also declined: Alphabet fell by 4% and Microsoft by 2%. Meanwhile, the stock markets have experienced significant downturns, particularly the Nasdaq, which dropped by 3.77%. On Monday the 27th, Dow Jones futures fell by 1.04%, while S&P 500 futures declined by 2.29%.

Photo: GPT4