Despite operating on a limited budget, Deepseek AI has positioned itself as a crucial tool and the top-ranked app in 2025 on the App Store and Play Store.
DeepSeek R-1 Release in January 2025 took the internet by storm, possibly hinting towards a replacement of ChatGPT sooner with this Chinese AI app.
Due to reported cyberattacks, DeepSeek is restricting new registrations, indicating potential security challenges, yet their rapid growth will blow your mind.
In this DeepSeek Statistics article, let’s check out the latest statistics, trends, and projections related to DeepSeek AI and help you understand where this platform is headed in 2025.
Top DeepSeek AI Statistics
- DeepSeek had about 7,475 visitors per day in August 2024. By February 2025, this grew to 100,776 visitors daily. That’s 12 times more people visiting their website. (Semrush)
- DeepSeek AI is trained on 14.8 trillion tokens. (DocsBot)
- It took 2.788 million H800 GPU hours to train DeepSeek. (DocsBot)
- DeepSeek used around 2,000 Nvidia H800 chips to train its models, showcasing substantial computational power. (DocsBot)
- The third version of DeepSeek (DeepSeek-V3) costs $5.5 million to create, which is only about 1/18th the cost of building OpenAI’s GPT-4. (Bussiness of Apps)
- Deepseek can remember up to 128,000 tokens at once, allowing it to process long pieces of text. In response, it can generate 8,000 tokens in one response. (Arxiv)
- DeepSeek-R1 outperforms OpenAI-o1-1217 in 2 out of 5 coding benchmarks (40%), leading in LiveCodeBench (65.9% vs. 63.4%) and SWE Verified (49.2% vs. 48.9%). (Arxiv)
How Many Users On DeepSeek AI?
- DeepSeek has amassed 2.4 million downloads on Google Play and 1.9 million downloads on the App Store, showcasing its strong presence across both platforms. (Backlinko)
- DeepSeek has over 10 million downloads as of January 2025. (Google Play)
- In just the first half month of January 2025, over 3 million people downloaded the DeepSeek AI app. (Bussiness of Apps)
DeepSeek App Users
- DeepSeek has gained significant traction among younger users, with 42.5% of iOS users and 50.4% of Android users falling within the 18-24 age group.
Let us have a look at the Deeepseek user demographics amongst iOS and Android users in detail:
Age Group | iOS Users | Android Users |
18-24 | 42.5% | 50.4% |
25-34 | 23.9% | 16.5% |
35-49 | 11.9% | 14.5% |
50-64 | 21.6% | 17.8% |
65+ | 0% | 0.8% |
(Backlinko)
- China leads iOS app downloads, accounting for 20.1% (391,765 downloads) of the total, followed by the United States at 16.2% (316,440 downloads).
Russia, Saudi Arabia, and Brazil also show strong adoption, highlighting DeepSeek’s growing global presence.
Here’s a detailed breakdown of Deepseek iOS downloads by country:
Country | Downloads | Share of Total |
China | 391,765 | 20.1% |
United States | 316,440 | 16.2% |
Russia | 117,963 | 6.0% |
Saudi Arabia | 66,505 | 3.4% |
Brazil | 64,046 | 3.3% |
United Kingdom | 56,515 | 2.9% |
Germany | 52,779 | 2.7% |
India | 50,975 | 2.6% |
France | 48,206 | 2.5% |
Turkey | 45,700 | 2.3% |
(Backlinko)
DeepSeek Performance Statistics
- DeepSeek started with about 7,475 daily visitors in August 2024.
- DeepSeek now gets 100,776 daily visitors (That is 12 times better than the performance from when it started).
The most dramatic growth occurred in January 2025, with traffic increasing by 312% compared to December 2024, coinciding with DeepSeek’s R1 release.
Here is a look at the monthly breakdown of Deepseek traffic value and average daily visitors:
Month | Avg Daily Visitors | Traffic Value ($) |
Aug 2024 | 12,473 | $6,571 |
Sept 2024 | 14,891 | $12,044 |
Oct 2024 | 14,753 | $14,302 |
Nov 2024 | 13,089 | $20,486 |
Dec 2024 | 17,825 | $31,246 |
Jan 2025 | 73,493 | $89,147 |
Feb 2025* | 97,193 | $103,459 |
*February data is partial (first week only)
Their website’s performance went from an average of 25,000 daily visitors to nearly 100,000 in just a few weeks in January 2025.
The value of DeepSeek’s traffic (if they had to pay for it through ads) grew from about $2,800 per day to over $100,000 per day in January 2025.
(Data from Semrush)
DeepSeek Downloads By Country
- China accounts for the largest portion of DeepSeek downloads, 39%, followed by the USA and France at 16% and 4%, respectively.
Here is the breakdown of DeepSeek downloads by country:
Country | Percentage of Downloads |
---|---|
China | 39% |
United States | 16% |
France | 4% |
Russia | 4% |
United Kingdom | 3% |
Saudi Arabia | 3% |
Germany | 3% |
Other Countries | 28% |
DeepSeek Important Facts
Here are some important facts about DeepSeek
Deepseek Cost | $0.14 per million input tokens $0.28 per million output tokens |
Founded On Date | May 1, 2023 |
Development Started | November 2023 |
Global Launch | January, 20th 2025 |
Deepseek Founder/CEO | Liang Wenfeng |
Country | Hangzhou, China |
Model | Multiple including DeepSeek-V3-Base, DeepSeek-R1-Zero, & DeepSeek-R1 |
Model Architecture | MoE (Mixture of Experts) |
Activated Parameters | 37 Billion |
Total Parameters | 671 Billion |
Company Size | ~200 employees (as of January 2025) |
Open Licensing | Open-source versions available |
Distilled Model Sizes | 1.5B, 7B, 8B, 14B, 32B, 70B |
Deepseek Funding | High-Flyer (Chinese hedge fund) |
(Arxiv, Deepseek, Bussiness of Apps)
Which Countries Have Banned DeepSeek?
Several countries have banned DeepSeek AI over security concerns. Italy removed it from app stores following a privacy probe.
Taiwan and Australia banned its use in government agencies. South Korea restricted it across key ministries.
The U.S. Congress, Navy, Pentagon, NASA, and Texas have also blocked or limited its use.
Find a detailed DeepSeek banned countries list with our article.
DeepSeek vs OpenAI Statistics & Data Comparison
DeepSeek performs on par with OpenAI-o1-1217. Here are the comparison details:
- DeepSeek matches or sometimes even exceeds OpenAI’s performance in mathematical reasoning tasks.
- AIME 2024: DeepSeek (79.8%) vs. OpenAI (79.2%)
- MATH-500: DeepSeek (97.3%) vs. OpenAI (96.4%) (Arxiv)
- DeepSeek-R1 performs better overall, but OpenAI leads in Codeforces Rating, Percentile, and Aider-Polyglot.
Here is a quick look at the numbers:
Benchmark | DeepSeek-R1 | OpenAI-o1-1217 | Winner |
---|---|---|---|
LiveCodeBench | 65.9% | 63.4% | DeepSeek |
Codeforces Rating | 2029 | 2061 | OpenAI |
Codeforces Percentile | 96.3% | 96.6% | OpenAI |
SWE Verified | 49.2% | 48.9% | DeepSeek |
Aider-Polyglot | 53.3% | 61.7% | OpenAI |
- DeepSeek-R1 outperforms OpenAI-o1-1217 in 2 out of 5 coding benchmarks (40%), leading in LiveCodeBench (65.9% vs. 63.4%) and SWE Verified (49.2% vs. 48.9%). (Arxiv)
- OpenAI-o1-1217 wins in 3 out of 5 benchmarks (60%), including Codeforces Rating (2061 vs. 2029), Codeforces Percentile (96.6% vs. 96.3%), and Aider-Polyglot (61.7% vs. 53.3%). (Arxiv)
- DeepSeek-R1 performs 4% better than OpenAI in LiveCodeBench (65.9% vs. 63.4%), indicating stronger general coding performance. (Arxiv)
- OpenAI leads by 3.9% in Aider-Polyglot (61.7% vs. 53.3%), suggesting better multilingual coding capabilities. (Arxiv)
- DeepSeek-R1 and OpenAI have near-identical performance in SWE Verified, with only a 0.3% difference (49.2% vs. 48.9%). (Arxiv)
DeepSeek Cost Comparison
- The development cost of DeepSeek is under $10 million, which is significantly lower than that of many other artificial intelligence companies with similar capabilities.
- DeepSeek-V3 has the lowest input token cost of $0.01, making it the most cost-efficient model. While Claude-3.5-Sonnet has the highest price at $30 per million tokens
Here is a simple breakdown of the input token costs for different AI models, listed from most expensive to least expensive:
Model | Input Token Cost per Million Tokens |
---|---|
Claude-3.5-Sonnet | $30.00 |
GPT-4 | $10.00 |
GPT-4 of-mini | $10.00 |
OpenAI GPT-4-mini | $7.00 |
Anthropic Claude-2.5-Turbo | $3.00 |
Google Gemini 1.5 Pro | $2.50 |
Meta Llama-3.5-70B | $2.00 |
Meta Llama-2.0-70B | $1.00 |
Meta Llama-2.0-110B | $0.70 |
Amazon Nova-SageMaker Pro | $0.30 |
OpenAI GPT-3.5 | $0.10 |
OpenAI GPT-4 Mini | $0.07 |
Mistral-Medium-Small-2k109 | $0.04 |
DeepSeek-V3 | $0.01 |
(DocsBot)
- DeepSeek models (DeepSeek-V3 and DeepSeek-R1) are among the cheapest, at $0.28 per million tokens.
- High-end models like OpenAI GPT-4, GPT-4-preview, and Claude-3.5-Sonnet charge $30 per million tokens generated.
Now, let us have a look at the Output token costs:
Model | Output Cost per Million Tokens |
---|---|
OpenAI GPT-4 | $30.00 |
OpenAI GPT-4-preview | $30.00 |
Claude-3.5-Sonnet | $30.00 |
OpenAI GPT-4-Turbo | $15.00 |
GPT-4 | $10.00 |
Google Gemini 1.5 Pro | $10.00 |
Google Gemini 1.5 Flash | $7.00 |
OpenAI GPT-4-mini | $3.00 |
Meta Llama-3.5-70B | $2.00 |
Meta Llama-2.0-70B | $1.00 |
Meta Llama-2.0-110B | $0.70 |
Amazon Nova-SageMaker Pro | $0.30 |
Cohere Command | $0.30 |
Mistral-Medium-Small-2k109 | $0.28 |
DeepSeek-V3 | $0.28 |
DeepSeek-R1 | $0.28 |
(DocsBot)
What’s The Big Deal About DeepSeek? (DeepSeek 2025 Contreversy)
DeepSeek first made waves by creating an AI that does not need the latest, most expensive chips but is as powerful as Google and OpenAI.
The big deal about DeepSeek was that it achieved this powerful AI with fewer resources and on a budget of under $10 million, less than its US competitors like Open AI and Google.
The end result of all this caused Nvidia’s stock to crash, wiping out $600 billion in market value (a 17% drop), leading to investors’ worry about the future demand for high-end chips.
Before the export ban, DeepSeek stockpiled 10,000 Nvidia A100 GPUs, securing long-term access to high-performance computing resources.
Can You Invest In DeepSeek Stock?
DeepSeek is a private company. Hence, you can not invest in it because it is not yet listed on the market.
DeepSeek does not appear to be publicly traded on stock exchanges, and there is no indication that it plans an initial public offering (IPO).
Future Of DeepSeek: Predictions & Possibilities
DeepSeek’s future seems bright due to its low cost and powerful capabilities. Its cost-effectiveness can help in healthcare, finance, customer service, and education.
However, DeepSeek must introduce advanced features and multimodal supports, such as text, voice, NLP (for humanization), and AI image generation.
If the company continues to lead in open AI development, it could play a pivotal role in setting standards for ethical AI practices, ensuring transparency and accessibility for everyone.
DeepSeek may also form partnerships with major tech companies using AI or government agencies, especially in regions like China and Russia, where AI development is prioritized. However, during global expansion, the company will face the major challenge of following local laws.
To get a deep understanding of what is happening in the field of AI, check our latest updated artificial intelligence statistics.