DeepSeek AI User Statistics and Facts (2025)
DeepSeek AI went live in January 2025 and became an instant hit, ranking #1 on the App Store in the top free app category, pushing ChatGPT to the 2nd position in the United States. Its app has been downloaded over 23 million times worldwide across all app stores within 19 days of launch, making it one of the most downloaded apps of all time.
The DeepSeek website attracted over 350.8 million visits worldwide in March 2026, of which 81.63% were from desktops and 18.37% from smartphones.
In this statistics blog, we have compiled the latest stats about DeepSeek AI and compared its performance with other major AI models in multiple benchmarks, providing a complete idea.
Top DeepSeek AI Statistics in 2026
- DeepSeek AI’s iOS and Android apps have been downloaded over 175 million times since their launch.
- DeepSeek supports over 80 programming languages.
- China contributes 34% of DeepSeek app downloads overall, and India accounts for 8% of total downloads.
- 60.48% of the platform’s users are male, while 39.52% are female.
- DeepSeek-R1 scored 79.8% in the AIME-2024 benchmark, surpassing OpenAI-o1-1217’s score of 79.2%.
- The input cost for the current flagship DeepSeek-V4-Pro is $0.435 per million tokens (cache miss), and the output cost is $0.87 per million tokens.
DeepSeek AI Important Facts
DeepSeek AI is new to the market, and most people are unaware of important facts like its parent company, employee count, founder, etc. To make it convenient, we have prepared a table with essential information about the chatbot.
| Metrics | DeepSeek Information |
| Launch Date | 1 May 2023 |
| Headquarters | Hangzhou, China |
| Founder | Liang Wenfeng |
| Parent Company | High-Flyer |
| Employees | Around 160 employees |
| Models | DeepSeek-V4-Pro, DeepSeek-V4-Flash, DeepSeek-V3.2, DeepSeek-R1 |
| Availability | Free |
| Model Architecture | Mixture of Experts (MoE) |
| DeepSeek Cost (V4 flagship) | $0.435 / million input tokens (cache miss)
$0.87 / million output tokens |
| Programming Language Support | Over 80 programming languages |
| Total Parameters (V-4 pro) | 1.6 trillion |
| Activated Parameters (V-4 pro) | 49 Billion |
Source: Deepseek documentation, LinkedIn.
What is DeepSeek AI?
DeepSeek AI is a free AI-powered chatbot developed by a Chinese company with the same name. Founded in July 2023, the platform specializes in open-source large language models (LLMs). The chatbot is developed with powerful yet cost-effective models like DeepSeek-V4-Pro and DeepSeek-V4-Flash, rivaling GPT-5.5 in efficiency. These sophisticated reasoning models are designed to handle complex reasoning tasks, including math, code, and general problem-solving.
DeepSeek AI is open-source and allows free use and modification of its AI algorithms. Since its app went live on the App Store and Play Store, it has pushed ChatGPT to the second position and ranked 1 on the App Store in the United States.
DeepSeek AI Timeline
The table below showcases the complete timeline of the key events of DeepSeek AI:
| Year | Month and Date | Key Events |
| 2023 | May | DeepSeek AI was founded by Liang Wenfeng in Hangzhou, China. |
| June | Launch of DeepSeek Coder and DeepSeek LLM (67B parameters) for coding and natural language processing. | |
| 2024 | January | Launch of DeepSeek LLM (67B parameters). |
| May | Launch of DeepSeek-V2 with enhanced performance and efficiency. | |
| Late 2024 | Launch of DeepSeek Coder-V2 (236B parameters) that offers a high context window (128K tokens). | |
| 2025 | January | Launch of DeepSeek-V3 (671B parameters) and DeepSeek-R1. |
| January 10 | Launch of DeepSeek chatbot app for iOS and Android, powered by DeepSeek-R1. | |
| January 27 | Becomes the most downloaded free app on the U.S. iOS App Store. | |
| March 25 | Release of DeepSeek-V3-0324 version with better front-end development skills. | |
| May 28 | Release of DeepSeek-R1-0528 update with improved benchmark performance and support for JSON output and function calling. | |
| August 21 | Release of DeepSeek-V3.1 update with hybrid reasoning architecture, faster reasoning compared to R1-0528, and 128K context for API. | |
| September 22 | Release of DeepSeek-V3.1-Terminus update with improved language consistency, Code Agent, and Search Agent functionalities. | |
| September 29 | Release of DeepSeek-V3.2-Exp update with introduction to DeepSeek Sparse Attention (DSA) for simple long-context training. This update also reduced API prices by over 50%. | |
| 2025 | December 1 | Release of DeepSeek-V3.2 alongside the DeepSeek-V3.2-Speciale variant focused on reasoning. |
| 2026 | April 24 | Release of DeepSeek-V4 preview, including DeepSeek-V4-Pro (1.6T parameters, 49B activated) and DeepSeek-V4-Flash (284B parameters, 13B activated). Both ship with a 1 million token context window and 384K max output, released under the MIT license. Legacy deepseek-chat and deepseek-reasoner aliases now map to V4-Flash. |
| 2026 | April | First external funding round opens at a $10 billion target valuation, climbing toward a reported $45 billion within weeks with Tencent and Alibaba reportedly in talks. |
| 2026 | July 24 | Scheduled deprecation of the legacy deepseek-chat and deepseek-reasoner API aliases (15:59 UTC), per the official DeepSeek API documentation. |
Source: DeepSeek AI
Number of DeepSeek AI Users
There are 130 million Deepseek AI users globally. The platform being local to China, the country leads in terms of the user count of the platform.
The DeepSeek app was downloaded over 23 million times across iOS and Android devices worldwide within 19 days of its launch on January 10, 2025. By the end of 2025, total downloads had reached approximately 175 million.
In contrast, the ChatGPT and Gemini apps were downloaded roughly 2.6 million and 783,632 times within their first launch month. According to the latest data, DeepSeek was ahead of ChatGPT and Gemini in terms of app downloads in its launch window.
DeepSeek AI Downloads by Country
Since the DeepSeek AI app launch on the App Store and Play Store, its download numbers have skyrocketed worldwide, with China, the United States, and India leading the list.
- 34% of the DeepSeek app downloads on iOS are from China, followed by India and the Russian Federation, with 8% and 7% respectively.
The table below shows the total number of DeepSeek AI app downloads by country (table based on 175 million downloads by the end of 2025)
| Country | Share of downloads | Number of downloads |
| China | 34% | 59 million |
| India | 8% | 14 million |
| Russia | 7% | 12 million |
| United States | 6% | 11 million |
| Pakistan | 4% | 7 million |
| Brazil | 4% | 7 million |
| Indonesia | 4% | 7 million |
| France | 3% | 5 million |
| United Kingdom | 3% | 5 million |
| Other | 27% |
Source: Statista, AppMagic, Google Play, Business of Apps, Backlinko
DeepSeek AI User Demographics Statistics
- Roughly 6 in 10 DeepSeek AI users are male, compared to 39.52% female.
- The 25 to 34 age group remains the largest cohort of DeepSeek’s web user base; on mobile, the 18 to 24 age group dominates, accounting for 44.9% of Android users and 38.7% of iOS users.
- Users aged 35 to 44 years account for 20.06% of the DeepSeek user base, while those aged 45 to 54 account for 13.21%.
- 4.87% of the platform's user base is aged over 65 years.
DeepSeek AI App Store Position Statistics
Within a week of launch, the DeepSeek AI app topped the United States App Store’s Top Free Category, placing the ChatGPT app second. An app needs over 200,000 daily downloads to rank first on the US Apple App Store list. This aligns with the DeepSeek AI app download statistics mentioned above.
- In June 2025, the DeepSeek app is in the 39th position in the App Store’s Top Free Apps list in around 60 countries worldwide.
- 80% of the app's downloads happened in the first 10 days of its launch, boosting its rank from 815th on January 19th to 1st on January 31st, 2025.
- At its peak in January 2026, DeepSeek ranked #1 on the App Store across over 156 countries, per Backlinko.
Source: AppMagic, Backlinko.
DeepSeek AI Google Trends Statistics
Google Trends data shows that DeepSeek search surged in the U.S. within 20 days of launch, reaching about 100 million searches. In comparison, ChatGPT recorded over 700 million searches during the same period.
In the United States, the District of Columbia tops the DeepSeek search list, followed by California, New Jersey, Washington, and Maryland. In comparison, Myanmar, Nepal, Ethiopia, Pakistan, and Kenya were the top 5 countries where DeepSeek AI was searched the most.
DeepSeek AI Traffic Statistics
As of March 2026, deepseek.com received over 350.8 million visits worldwide, a 6.69% increase from the prior month. The website ranks 4th in the AI Chatbots and Tools category and 118th worldwide.
- DeepSeek AI’s traffic is 81.63% from desktops and 18.37% from mobile web.
- The website’s average daily traffic is roughly 11.3 million, and an average session lasts around 5 minutes and 2 seconds.
- The bounce rate of DeepSeek AI’s website is about 38.08%.
- China contributes roughly 44.17% of deepseek.com’s total traffic, the highest of all countries.
- Russia and the U.S. rank second and third, accounting for 10.35% and 5.68% of traffic, respectively.
- Hong Kong ranks fourth with a 3.65% traffic share, while Brazil ranks fifth, contributing 3.19%.
- 73.04% of deepseek.com’s desktop traffic comes from direct visits, while the remainder is split between organic search and referrals.
- The top organic search terms for deepseek.com are deepseek, deep seek, deepseek r1, deepseek ai, and deepseek v3.
- YouTube contributes around 59.79% of the traffic share of DeepSeek’s website, followed by WhatsApp and Facebook with 10.97% and 6.31%, respectively.
- VKontakte and Telegram Webapp rank fourth and fifth, accounting for 5.49% and 3.34% of the traffic share, respectively.
Source: SimilarWeb
DeepSeek AI vs. OpenAI Comparison Statistics
Studies conducted at DeepSeek-R1’s January 2025 launch showed that the DeepSeek-R1 model performed similarly to the OpenAI-o1-1217 model in reasoning tasks. The DeepSeek-R1-Zero model, trained through large-scale reinforcement, showcased impressive reasoning skills and powerful, intriguing behaviors. DeepSeek demonstrated comparable performance to OpenAI, often surpassing it in certain scenarios. Here are some interesting statistics data showcasing the comparison between DeepSeek and OpenAI in different benchmarks.
- In the AIME-2024 performance, DeepSeek-R1 scored 79.8%, slightly higher than 79.2% of OpenAI-o1-1217. DeepSeek-R1-32B scored 72.6%, while OpenAI-o1-mini, and DeepSeek-V3 scored 63.6% and 39.2%, respectively.
- In the Codeforces performance, DeepSeek-R1 and OpenAI-o1-1217 performed on par with 96.3% and 96.6% scores, respectively.
- In the GPQA Diamond performance, OpenAI-o1-1217 scored 75.7%, performing better than DeepSeek-R1 with 71.5%.
- In the MATH-500 performance, DeepSeek-R1 scored 97.3%, higher than OpenAI-o1-1217 with 96.4%.
- In the MMLU performance, OpenAI-o1-1217 scored 91.8%, slightly higher than DeepSeek-R1 with 90.8%.
- In the SWE-bench Verified performance, DeepSeek-R1 and OpenAI-o1-1217 performed similarly with 49.2% and 48.9%, respectively.
The table below shows the benchmark performance data of DeepSeek-R1 vs. OpenAI-o1-1217 vs. DeepSeek-R1-32B vs. OpenAI-o1-mini vs. DeepSeek-V3.
| Benchmarks | DeepSeek-R1 | OpenAI-o1-1217 | DeepSeek-R1-32B | OpenAI-o1-mini | DeepSeek-V3 | Winner |
| AIME-2024 | 79.8% | 79.2% | 72.6% | 63.6% | 39.2% | DeepSeek |
| Codeforces | 96.3% | 96.6% | 90.6% | 93.4% | 58.7% | OpenAI |
| GPQA Diamond | 71.5% | 75.7% | 62.1% | 60% | 59.1% | OpenAI |
| MATH-500 | 97.3% | 96.4% | 94.3% | 90% | 90.2% | DeepSeek |
| MMLU | 90.8% | 91.8% | 87.4% | 85.2% | 88.5% | OpenAI |
| SWE-bench Verified | 49.2% | 48.9% | 36.8% | 41.6% | 42% | DeepSeek |
Source: Arxiv
DeepSeek vs. Gemini vs. Qwen vs. GPT4o Benchmark Comparison Statistics
Like DeepSeek, Qwen is another large language model launched by another Chinese brand, Alibaba Cloud. It was launched in April 2023 and handles multiple tasks, such as natural language understanding, generation, and visual language processing.
Qwen is built on Meta’s Llama model and has evolved with various updates, including Qwen 2.5-Max, launched in January 2025. Like DeepSeek, it offers open-source versions, allowing developers to modify and use them for multiple tasks.
Below, we have compared the january 2025 benchmark scores of DeepSeek, Gemini, Qwen, and GPT4o models.
| Benchmarks | DeepSeek-R1 | Gemini 1.5 Pro (002) | Qwen2.5-Max | GPT-4o |
| MMLU | 90.8% | NA | 87.9% | 88.7% |
| MMLU-Pro | 84% | 75.8% | 69% | 74.68% |
| BBH | NA | NA | 89.3% | NA |
| C-Eval | NA | NA | 92.2% | NA |
| CMMLU | NA | NA | 91.9% | NA |
| MMMU | NA | 65.9% | NA | 69.1% |
| HellaSwag | NA | 93.3% | NA | NA |
| HumanEval | NA | 84.1% | 73.2% | 90.2% |
| MBPP | NA | NA | 80.6% | NA |
| CRUX-I | NA | NA | 70.1% | NA |
| CRUX-O | NA | NA | 79.1% | NA |
| MATH | 97.5% | 86.5% | 68.5% | 75.9% |
| GPQA | 71.5% | 59.1% | NA | 53.6% |
| GSM8K | NA | NA | 94.5% | NA |
| IFEval | 83.3% | NA | NA | NA |
- In the MMLU-Pro benchmark, DeepSeek-R1 scored 84%, higher than Qwen2.5-Max and GPT-4o.
- Qwe2.5-Max scored 89.4% in the Arena-Hard benchmark, higher than DeepSeek-V3 and GPT-4o.
- In the LiveBench benchmark, Qwe2.5-Max scored 62.2%, higher than DeepSeek-V3 and GPT-4o.
Source: Qwen, DocBot 1, DocsBot 2, DocsBot 3
DeepSeek AI Cost Comparison Against Other AI Models
According to the official DeepSeek API documentation as of June 2026, the input cost for processing tokens of the flagship DeepSeek-V4-Pro is $0.435 per million tokens (cache miss). Its output cost for tokens generated by the model is $0.87 per million tokens. The cheaper DeepSeek-V4-Flash variant is priced even lower at $0.14 per million input tokens and $0.28 per million output tokens.
In comparison, the input cost for processing tokens of OpenAI’s GPT-5.5 is $5.00 per million tokens, while the output cost is $30.00 per million tokens. Whereas, the price for similar tasks of Anthropic’s Claude Opus 4.7 is $5.00 per million tokens (input cost) and $25.00 per million tokens (output cost).
DeepSeek-V4-Pro has a significantly lower input and output token cost than GPT-5.5 and Claude Opus 4.7. The table below shows the price of input and output tokens of the top AI models as of June 2026:
| Models | Input Token Cost (Per Million Tokens) | Output Token Cost (Per Million Tokens) |
| OpenAI GPT-5.5 | $5.00 | $30.00 |
| Claude Opus 4.7 | $5.00 | $25.00 |
| Claude Sonnet 4.6 | $3.00 | $15.00 |
| Gemini 2.5 Pro | $1.25 | $10.00 |
| Amazon Nova Pro | $0.80 | $3.20 |
| Meta Llama 4 (405B) | $1.79 | $1.79 |
| DeepSeek-V4-Pro | $0.435 | $0.87 |
| DeepSeek-V4-Flash | $0.14 | $0.28 |
Source: DeepSeek API documentation, CloudZero, official model pricing pages.
Impact of DeepSeek Launch in the United States Stock Market
The launch of DeepSeek triggered a sharp selloff in the U.S. stock market, particularly in the tech sector. Here are some interesting statistical points you should know:
- Since the launch of the DeepSeek AI app, the US market has lost $1 trillion, with top tech companies like Nvidia, Microsoft, and Google facing major declines.
- Nvidia suffered a massive loss from this launch, with its stock plunging 17%, wiping out nearly $600 billion in market value. Whereas Microsoft‘s stock fell by 2.14%, while Google‘s dropped by 4%.
- The SandP 500 tech sector also dropped by a staggering 5.6%, marking its biggest decline since September 2020.
Despite initial criticism, the wider availability of efficient AI models like DeepSeek is set to drive market growth, boosting innovation and productivity across sectors in the medium term.
Source: MSN, U.S.News, CBS News, Yahoo Finance
Drawbacks and Controversies of DeepSeek in 2025-2026
DeepSeek's launch has sparked controversy due to censorship, data privacy concerns, and disinformation risks. Here are some controversies and concerns of the platform you must know:
- Censorship and Bias: DeepSeek has faced criticism for aligning with Chinese government narratives. It avoids discussing sensitive topics like the Tiananmen protests and the Cultural Revolution, which raises concerns about censorship and biased responses.
- Data Privacy Issues: DeepSeek collects extensive user data, including keystroke patterns for biometric identification. This level of data collection is seen as intrusive, sparking concerns over privacy and security.
- Disinformation Risks: Experts warn that DeepSeek could be used to spread disinformation, undermine public trust, and reinforce authoritarian narratives, particularly in democratic societies.
Conclusion (Future Predictions of DeepSeek)
The launch of DeepSeek has undoubtedly disrupted the AI industry. With major tech companies losing billions of dollars, this shows how disruptive it was. The company challenges major players like OpenAI and Google with its cost-effective and efficient models. Its open-source approach has made AI accessible to all. With Alibaba launching Qwen, it is predicted that various Chinese and other countries' companies will launch AI models in the future, ending the dominance of US tech giants. However, concerns about data privacy and misuse persist with Chinese platforms.
FAQs
1) Is DeepSeek better than ChatGPT?
The comparison between DeepSeek and ChatGPT depends on specific use cases. According to benchmark performance, DeepSeek is better than ChatGPT in some scenarios. However, the major difference is in the operating cost. DeepSeek is much cheaper to operate than ChatGPT.
2) Who is the owner of DeepSeek?
DeepSeek is owned and funded by the Chinese hedge fund High-Flyer. The company was founded by Liang Wenfeng in 2023.
3) Is DeepSeek free?
DeepSeek offers free access to its AI models, like DeepSeek-V3 and R1. You can access its services for free through its official website, and the DeepSeek app is available for free download on both the Apple App Store and Google Play Store
4) What chips does DeepSeek use?
DeepSeek primarily uses NVIDIA GPUs for its models. Due to the chip ban, it uses Nvidia’s H800 chips, which it could have purchased legally in 2023. The brand also has access to Nvidia’s less powerful H20 chips, and as of 2026 has also been encouraged by Chinese authorities to adopt Huawei Ascend chips for inference workloads.
