After ChatGPT became a global sensation in 2022, the next big moment in AI came in early 2025 with DeepSeek. This relatively unknown Chinese company surprised everyone by challenging big players like OpenAI and Anthropic.
DeepSeek AI is an AI chatbot similar to ChatGPT. It was created by Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., a company owned by the Chinese hedge fund High-Flyer. The founder and CEO is Liang Wenfeng, who also co-founded High-Flyer.
DeepSeek gained fame in January 2025 when it released the R1 reasoning model. It nearly matched OpenAI’s o1 model and even surpassed ChatGPT on the US App Store. What made DeepSeek special was its low training cost. The R1 model reportedly cost just $6 million to train, far below the $100 million often spent on Western AI models. This move even caused Nvidia’s share price to drop. DeepSeek also shared the model weights publicly, pushing for open-source AI development.
The company published a detailed technical paper explaining its RL-based post-training method. Unlike OpenAI, DeepSeek openly shared how it trained its reasoning models. When OpenAI charged $20 for o1 access, DeepSeek released R1 for free. This helped make advanced AI more accessible worldwide.
DeepSeek has released several AI models over the years. In 2023, it launched DeepSeek Coder for coding, DeepSeek LLM, and DeepSeek-Math. Later versions included DeepSeek-V2, DeepSeek-Coder V2, DeepSeek-V2.5, DeepSeek-R1-Lite, DeepSeek-V3, and the main DeepSeek-R1 reasoning model in January 2025. The latest versions, DeepSeek-R1-0528 and DeepSeek-V3.2, came out in September 2025. More big models are expected soon.
You can use DeepSeek AI for free on chat.deepseek.com or via the mobile app on Android and iOS. You can also run smaller versions of R1 locally on your device.
Recently, DeepSeek has slowed down on new launches. This is because the Chinese government asked the company to use domestic AI chips, like Huawei Ascend, for its next model. DeepSeek tried using Ascend chips but ran into technical limits and returned to Nvidia GPUs for training, though it uses Ascend chips for inference.
DeepSeek has quickly become one of China’s top AI companies alongside Alibaba. Its rise shows that powerful AI models can be built at a lower cost and made more accessible to users worldwide.
