2025 Rookie DeepSeek: Facts, Statistics, and Key Insights

In January 2025, a new conversational AI tool, DeepSeek, was launched. Since its release, it has quickly topped the open-source modeling charts in just one day.

DeepSeek-R1 has garnered global attention with performance comparable to OpenAI's GPT-4. Many expect it to become the top-ranked app on the App Store and Play Store in 2025.

In this DeepSeek R1 blog, you'll learn more about its breakthroughs, achievements, controversies, and future prospects.

DeepSeek, full name Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd, is an innovative technology company founded on July 17, 2023, focusing on the development of advanced Large Language Models (LLMs) and related technologies.

Key Milestones:

• January 2024: Released DeepSeek LLM (first-generation model).
• May 2024: Open-sourced DeepSeek-V2, a second-generation MoE model.
• January 2025: Launched DeepSeek R1, with performance comparable to OpenAI's O1 version.

It is worth noting that DeepSeek R1 has garnered global attention, ranking among the world’s leading AI models. It has not only delivered outstanding performance in international AI model ranking competitions, but its application has also topped the free charts on the Apple App Store in both China and the United States.

Here are some important facts about DeepSeek company.

Source: DemandSage

The most talked-about aspect of the discussion around DeepSeek AI is its innovation. Whether it’s in terms of tokens or parameters such as GPU hours, it has played a major role in advancing the AI field, setting a new standard for both efficiency and cost-effectiveness.

• DeepSeek V3 used about 671 billion parameters and 14.8 trillion tokens.



• DeepSeek has open-sourced its flagship model as well as six smaller variants ranging from 1.5 to 70 billion parameters.
• DeepSeek achieved the benchmark using only 2.8 million H800 GPU hours of training hardware time (equivalent to approximately 4e24 FLOPs).
• DeepSeek uses about 2,000 Nvidia H800 chips to train its model, demonstrating powerful computational capabilities.
• DeepSeek can remember up to 128,000 tokens at once for processing long texts and generate up to 8,000 tokens in a single response.

Source: StatsUp

As of now, DeepSeek has been having a significant global impact, attracting millions of users to search and engage. Since its global launch on January 20, 2025, it has maintained an average of 1.8 million daily active users.

Source: StatsUp

On average, each visitor spends about 5 minutes and 25 seconds per visit, browsing roughly 3.5 pages. Meanwhile, DeepSeek sees a bounce rate of 36%, meaning that over a third of visitors leave after viewing just one page.

Among DeepSeek's all visitors, 71.57% are male, while 28.43% are female.

Source: StatsUp

The largest user age group is 25-34, making up 34.47% of the audience, followed by the 18-24 group at 22.3% and the 35-44 group at 19.82%.

Source: StatsUp

"Deepseek" was the most searched keyword among all users, generating a total of 1.7 million searches and bringing in the most search traffic to the site.

Source: StatsUp

As a result of DeepSeek's tremendous traffic, there were 364,288 tweets related to DeepSeek as of January 29th.

As of December 2024, DeepSeek's website had received 11.8 million visits, with direct traffic making up 61.54% of the total. Nearly a third of the traffic (31.93%) came from organic search, highlighting DeepSeek’s reliance on search engines. (Date source: StatsUp)

The high volume of traffic has also led to a high volume of downloads, with more than 10 million downloads of DeepSeek as of January 2025, meaning that more than 3 million people downloaded the DeepSeek AI app in the first half of January 2025 alone.

Of the total downloads, 39% came from China, with the US at 16%, India at 10%, and South Korea at 7%.

Source: AppMagic

To get a clearer picture of DeepSeek's innovations, here's a data visualization to show how it compares to OpenAI in terms of benchmarks and cost.

DeepSeek AI's performance in mathematical reasoning tasks is comparable to OpenIAI.

Source: Arxiv

• AIME 2024: DeepSeek (79.8%) vs. OpenAI (79.2%)
• MATH-500: DeepSeek (97.3%) vs. OpenAI (96.4%)

Other than that, in terms of other benchmarks, DeepSeek AI and OpenAI are neck-and-neck, with each having better-performing data, as shown in the following comparisons.

Source: Arxiv

• DeepSeek-R1 outperforms OpenAI by 4% on LiveCodeBench, indicating that DeepSeek R1 has better general-purpose coding performance.
• OpenAI's 3.9% lead in Aider-Polyglot demonstrates its multilingual coding ability over DeepSeek R1.

After evaluating the performance of the two AI tools in real-world applications, the next step is to analyze their respective costs. Here are the statistics for DeepSeek V3 and OpenAI GPT-4.

Source: Company data

• The DeepSeek models (DeepSeek-V3 and DeepSeek-R1) are among the cheapest models, selling for $0.28 per million tokens.
• High-end models such as OpenAI GPT-4, GPT-4-preview, and Claude-3.5-Sonnet charge $30 per million tokens generated. (Data source: DemandSage)

Overall, developing DeepSeek costs less than $10 million, significantly lower than many other AI companies with similar capabilities. With an input token cost as low as $0.01, DeepSeek-V3 is the most cost-effective model available.

As DeepSeek has attracted global attention and gained significant exposure, it has inevitably been targeted by hackers, raising concerns about its security.

On January 29, cybersecurity researchers at Wiz Research revealed that DeepSeek had suffered a major data breach, compromising over a million sensitive records. However, DeepSeek had already addressed the vulnerability before Wiz released its findings. Additionally, according to Palo Alto Networks, DeepSeek's model is "fairly easy" to crack, potentially aiding in the development of code for hacking, phishing, or social engineering attacks. (Date source: Wiz)

However, these challenges are not unique to DeepSeek. Other AI tools have faced similar dilemmas during their development. Despite these issues, DeepSeek's low cost and powerful capabilities far outweigh its shortcomings. If it continues to advance its AI technology and maintain its competitive edge, the future looks promising across sectors such as healthcare, finance, customer service, and education, positioning DeepSeek to lead the AI industry.

Was this page helpful?

Thanks for your rating

Rated successfully!

You have already rated this article, please do not repeat scoring!

Article by

Heather Marston

With her strong interest in technology related to mobile data, Heather Marston devotes herself to writing technical articles and sharing her experience using Apple and Android devices in a better way.