DeepSeek, a Chinese AI system that was developed in China by the year 2025, has been gaining traction. The Chinese AI model DeepSeek’s rise sparked a crash in the stock market for technology and almost caused heart attacks to top project managers.

DeepSeek is a search engine.

DeepSeek, a Chinese firm, specializes in artificial intelligence and big data technology. DeepSeek’s primary focus is to create solutions for machine learning, data analysis and process automation. DeepSeek develops platforms and tools that enable businesses to use data effectively for improving operations and products, making better decisions, and optimizing their decision-making processes.

DeepSeek also works in the areas of natural language processing, computer vision and other AI fields to develop innovative products, such as analytical platforms, intelligent assistants and recommendation systems. They apply their technologies to a variety of industries including retail, finance, health care, and logistics.

DeepSeek is the company’s chatbot. The AI program uses NLP. This program is both a browser-based application and a mobile app. It is available in any language.

What the bot says about itself

I am an artificially intelligent system created by DeepSeek. I am here to answer your questions, give you information, help with tasks and have a dialogue about various topics. “I use natural language processing and machine learning to generate answers and understand requests.”

DeepSeek was launched in 2023. The first chatbot version was created that year. DeepSeek didn’t gain widespread popularity until January 2025.

The following is a brief introduction to the topic:

Architecture

DeepSeek V3 uses a Multi-head Attention architecture (MLA), which allows for efficient inference, and achieves performance that is comparable with leading closed-source model. This model uses a DeepSeekMoE Architecture, which encourages and facilitates economic training. MLA is designed for efficient inference, reducing computational complexity. DeepSeekMoE enables cost-effective and effective training through a reduction of parameters.

Train the Trainer

DeepSeek-V3’s training framework is designed for efficient inference, and to provide cost-effective instruction. The framework adopts a full computation-communication overlap strategy, which enables the model to achieve efficient inference and reduces training costs. The framework employs an cross-node MoE strategy that enables models to be trained efficiently and at low cost on large-scale models.

DeepSeek Version 3: How to Use it

Create an account to start using DeepSeek. You will need to enter your email address, and create a password. A verification code will be sent to the address specified by the program.

Users can start working on the AI model immediately without having to pay anything.

ChatGPT’s main competitor in large language models

The Chatbot Market was revolutionized by American OpenAI’s ChatGPT. Before it, no other software program had such features.

OpenAI first released ChatGPT as a beta on November 30th, 2022. It was based off the GPT-3.5 Model. ChatGPT plus with GPT-4 will be released in March 2023. The chatbot says that it has been evolving since then and is constantly improving its abilities.

ChatGPT will offer several different models in the web and app versions by January 2025. Most advanced model is the o1(omni). Developers claim that it is a significant improvement over its predecessors.

A new model named DeepSeek R1 was launched on January 20th, 2025. The model is based on the outdated deepseek version 3. This version also contains elements of deepseek coder V2 (this neural network is designed to perform coding tasks).

DeepSeek R1’s release caused panic on the stock exchange and a drop in AI stock prices. The release also had a negative impact on cryptocurrency prices. A battle between two titans began that very day: Chinese DeepSeek versus American OpenAI.

DeepSeek is cheaper than ChatGPT. It costs less to develop and maintain the chatbot. The subscription fee for the Chinese neural networks is also lower. Here are the rates for ChatGPT and here for DeepSeek.

Chinese experts were able to reduce costs in part by using a “mixture” of expertise. It is an approach to working with data that separates the processing and applies a different expert for every task. The scheme saved developers time by allowing them to train the model quickly and efficiently. It also allowed for the organization of work across a group of NVIDIA graphics cards that were relatively cheap. The project team claims that they have spent 20 times less than competitors on the creation and maintenance of their language models.

The “Chinese”, too, has an open-source source code. Anyone can now customize the neural networks to suit their needs. ChatGPT users could not have imagined such a possibility.

What is the cost of DeepSeek?

The official company website lists the following rates:

This is how you calculate costs (spoiler alert: It’s complex):

1. Calculate the number of tokens

The input tokens are the values you provide to the model.

The model generates a text output token.

You need to take into account both the tokens that were used for the CoT and the tokens you use for the answer.

2. Viral level today = Yellow per token

DarkSalmon’s Today’s Viral level depends on model and token count.

If the price for 1,000,000 tokens was $0.14 then the cost for 10,000 tokens would be: 10,000 tokens x (0.14/1,000,000 tokens = $0.0014

3. Special Offers

All models, except DeepSeek R1, are eligible for discounts until 8 February 2025. Prices will then return to normal after that date.

4. Payment Processing

The funds are taken from your account (if you have any balance). Deductions are based on how many tokens you have used.

Cost Calculation

The cost is the number of tokens multiplied by today’s viral level.

Let’s say we want to create an article using the Chinese model, which ultimately has 10,000 words. This is how you would calculate:

For us to calculate the price, we must know how many tokens are required per 10,000 words.

1. How many tokens are there in 10,000 words?

In general, one word equals about 1.3 tokens. Then, 10,000 tokens would equal approximately 13,000 words.

2. Tomatoes for tokens

If the price of 1,000,000 tokens (as shown in the example) is $0.14, then the answer to the question Where can I buy them? will be:

3. Calculate Cost

The cost is:

13,000 tokens x (0.14 /1 000 000) = $0.00182.

The cost of generating 10,000 words is approximately $0.00182. The exact price depends on which model was used to generate the text.

Interesting Fact: In order to not lose users due to the success of the Chinese alternatives, OpenAI introduced promo codes as well as other discounts.

The payment is calculated differently by ChatGPT

The chatbot is charged according to the monthly tariff or the annual tariff. Tariff plans are available based on the needs of your business.

ChatGPT (version o1) itself admits, and this is worth noting: its rival offers far more favorable conditions of employment.

ChatGPT and DeepSeek: Conflict between the two

OpenAI claims DeepSeek has used its model for training their chatbot. The statement raises concerns over possible intellectual property rights violations.

OpenAI claims that they have found evidence of the use of distillation techniques. This is a way for developers to improve performance by using insight from more powerful, larger models. It is possible to achieve similar results for specific tasks with much less cost.

OpenAI did not disclose the details of their evidence, but stated that it’s terms of service states that users can’t “copy” OpenAI services or use its outputs to create competing models.

OpenAI claims were found amusing by users on social media.

Sorry, I just can’t help myself. OpenAI is upset that DeepSeek might have used outputs of ChatGPT to train. This company was built by stealing the internet.

There are also reports that DeepSeek was initially presenting itself as ChatGPT. Those who posted about this issue claim that developers fixed the bug quickly. Memes quickly grew out of the conflict between ChatGPT vs DeepSeek.

OpenAI to DeepSeek : “You’re trying steal my stolen data.”

Also, jokes were made about ChatGPT’s reaction to a new competitor.

DeepSeek’s rapid entry into the battle between ChatGPT, Google Gemini and DeepSeek was a source of amusement for users.

For a very long time, the teams behind these neural networks were competing to be the market leaders. Something went wrong with DeepSeek.

DeepSeek under fire

The confrontation between DeepSeek vs. ChatGPT is seen by many as another battle in the political arena between China and the USA. There have been reports about the potential of a ban in America on Chinese neural networks. Media reports that Congress is urging Trump on new restrictions for Nvidia chips, which are used in DeepSeek.

Bloomberg reports that just days after China’s latest artificial intelligence model was released, hundreds of businesses began to take steps to ban it. There are also reports that Microsoft and OpenAI have been investigating DeepSeek.

The Chinese neural network was already banned in Italy. Data protection authorities in Italy banned DeepSeek because of a lack information about how the software uses personal data. French and Irish regulators are also looking at the neural network.

Which is better: DeepSeek or ChatGPT in terms of reasoning abilities?

There are differing opinions on which is superior, DeepSeek or ChatGPT. Others believe the Chinese neural networks far outperform their American rivals, while some think that DeepSeek is still a ways off, no matter how much ChatGPT costs.

Professor Mushtaqbilal has compared both models to see how effective they are. He is an expert in ethical applications of artificial intelligence to academic ends. He presented his findings in the form of a graph.

How to Read it

This graph displays accuracy and percentiles across five AI tests: DeepSeek R1, OpenAI o1-1217 DeepSeek R1-32B OpenAI o1 mini and DeepSeek V3. The names of these tests are:

AIME 2024: A test that measures the percent of correct answers.

Codeforces Percentile: An evaluation based on competition results on the Codeforces Platform.

GPQA Diamond: A test to solve questions at the first try

Test focusing on mathematical problem solving.

Multitasking Test (MMLU): Pass@1

The SWE Bench Verified Test (Resolved), a test to solve software development problems.

According to the results of this test, DeepSeek R1 is superior in terms correctness. ChatGPT 01 was the winner of the Codeforces Test. The ChatGPT 01 also beats its rival in terms of solving problems on the very first try. DeepSeek is unbeatable when it comes solving math problems. In addition to programming tasks, the Chinese neural network outperforms its competitor. ChatGPT, on the other hand excels at multitasking situations.

What else are users saying about DeepSeek?

As the buzz about DeepSeek was just beginning to die down, news broke that Alibaba’s Qwen artificial Intelligence had a rival. In the best tradition of marketing, company representatives claim their product to be more efficient than competitors. There are other Chinese-language models which could also join the race.

Considerations Instead of Conclusions

Users have started to discuss the beginning of a brand new era in language models with the advent of DeepSeek. Developers will do anything to beat their competition in this race. OpenAI’s response to a Chinese competitor is an example of this. Tests have shown that DeepSeek has already surpassed ChatGPT on several metrics. The Chinese neural network has some drawbacks. Its accessibility, however, outweighs these limitations.

The post What You Should Know about DeepSeek, and how it differs from ChatGPT. appeared first on Crypto Giggle. Your home for bite sized Crypto Newsthat won’t put you to sleep!

---

Filed under: News - @ February 2, 2025 7:09 am