On January 20, 2025, a Chinese research lab made a groundbreaking move in the world of artificial intelligence by launching DeepSeek R1. This AI chatbot quickly gained global attention, outperforming top models from OpenAI, Meta, and Google in benchmarks like mathematics and reasoning. DeepSeek R1’s remarkable efficiency and superior performance have set a new standard, disrupting the AI industry in ways no one anticipated.
DeepSeek vs. Western AI Models
What sets DeepSeek apart is not just its performance but also its cost-efficiency. While OpenAI charges $200 per month for its ChatGPT o1 Pro model, DeepSeek is completely free for users. Moreover, the development cost of DeepSeek was just $5.6 million— a fraction of the billions spent by American tech giants to develop their AI models.
This innovation has also impacted the stock market. Following its release, DeepSeek became the most downloaded app in the United States on both the App Store and Google Play, surpassing ChatGPT. By January 27, 2025, its influence had rippled through the financial sector, wiping out nearly $1 trillion from the U.S. tech market. NVIDIA, the world’s most valuable company before DeepSeek’s launch, saw its market capitalization plummet from $3.5 trillion to $2.9 trillion in a single day, losing $589 billion—the largest loss any company has ever faced.
Who Created DeepSeek?
The mastermind behind DeepSeek is 40-year-old Chinese entrepreneur Liang Wenfeng. Known for his reclusive nature, Wenfeng rarely makes public appearances, and little is known about his personal history. However, his background in AI and finance is notable. In 2015, he founded a hedge fund named High Flyer, which utilized AI for investments. Four years later, he established High Flyer AI, focusing on artificial intelligence research.
By May 2023, Wenfeng, driven by scientific curiosity rather than financial gain, funneled his hedge fund earnings into the development of an AI model. Instead of hiring engineers, he built a team of PhD students from China’s top universities. After just two years and minimal funding, DeepSeek R1 was ready, shocking the world with its capabilities.
How DeepSeek Works: Chain of Thought Processing
DeepSeek’s architecture is built on the ‘Chain of Thought’ model, similar to OpenAI’s ChatGPT o1. This method enhances logical reasoning by allowing the AI to break down complex problems step by step before arriving at an answer.
For example, when asked, “Which is larger, 9.11 or 9.9?” older AI models might instantly reply that 9.11 is larger, which is incorrect. DeepSeek, however, pauses to reassess its answer, counter-questions itself, and ultimately delivers the correct response: 9.9 is greater.
Unlike previous models, DeepSeek openly displays its thought process. When given a task, it carefully evaluates different perspectives before generating a final response. This approach mimics human-like reasoning and has set a new benchmark for AI capabilities.
Performance Comparisons: How DeepSeek Stacks Up
AI experts have compared DeepSeek with leading AI chatbots such as OpenAI’s ChatGPT, Google’s Gemini, and Meta’s Llama. In key areas, DeepSeek has emerged as a leader:
- Coding: DeepSeek ranks highest, followed by ChatGPT and Claude.
- Quantitative reasoning: DeepSeek outperforms ChatGPT and Qwen 2.5.
- Scientific knowledge: ChatGPT leads slightly, but DeepSeek follows closely.
- Riddle-solving: DeepSeek excels, proving its advanced reasoning skills.
However, DeepSeek has a major drawback—response time. While ChatGPT o1 takes an average of 31.15 seconds to generate a response, DeepSeek requires 71.22 seconds. This latency is increasing due to the high demand for the model, causing server overloads.
DeepSeek’s Biggest Drawback: Censorship
Despite its impressive capabilities, DeepSeek has faced criticism for its censorship policies. When asked politically sensitive questions—such as topics related to Tiananmen Square, Taiwan’s independence, or criticism of Xi Jinping—DeepSeek simply responds, “Sorry, I am not sure how to approach this type of question yet.”
However, when asked to critique global leaders like Joe Biden, Donald Trump, or Vladimir Putin, DeepSeek provides detailed responses. This selective censorship is enforced by China’s Cyberspace Administration, which rigorously tests AI models with 70,000 politically sensitive questions before approving their public release.
The Open-Source Advantage
Despite its censorship issues, DeepSeek has gained widespread adoption due to its open-source nature. Unlike OpenAI, which started as an open-source initiative but later restricted access, DeepSeek’s code is freely available. This allows users and companies to modify the model for their needs.
Perplexity AI, for instance, downloaded DeepSeek R1, removed its censorship restrictions, and integrated it into their platform. Microsoft also adopted DeepSeek R1, making it available on Azure AI Foundry, with plans to integrate it into its Copilot assistant.
Innovations in AI: The Mixture of Experts Method
One of DeepSeek’s most innovative features is its ‘Mixture of Experts’ model. While ChatGPT o1 operates as a single model handling all types of queries, DeepSeek divides tasks among specialized sub-models. It assigns questions related to engineering, medicine, or law to dedicated expert models, improving accuracy and reducing data transfer time.
Final Thoughts
DeepSeek R1 has taken the AI world by storm, proving that AI innovation is no longer limited to the West. With its open-source framework, superior reasoning capabilities, and cost-efficiency, DeepSeek has set a new benchmark for AI development. However, its censorship policies remain a major concern.
As AI continues to evolve, DeepSeek’s emergence signals a shift in the global AI landscape—one where China is increasingly becoming a dominant player. Whether DeepSeek can maintain its momentum in the face of competition remains to be seen, but one thing is clear: the AI race has never been more intense.
