DeepSeek is a Chinese AI company, as well as the name of its model family and chatbot. It made headlines in early 2025 by releasing a state-of-the-art reasoning model that shook things up in Silicon Valley and knocked a few billion off major tech company stocks.
Things have settled down since then, and DeepSeek is now one of a number of major Chinese AI players, but it's definitely still worth talking about.
Table of contents:
What is DeepSeek?
DeepSeek is a family of AI models built by DeepSeek, the Chinese AI company. It's also the name of the chatbot and API platform that uses the model family. I'm going to touch on all three instances of DeepSeek in this article, but for now, let's focus on the model family.
DeepSeek V3 is DeepSeek's flagship open model; it powers the DeepSeek chatbot and is the only model available through the official DeepSeek API platform. It's currently on V3.2, though there are rumors it will be replaced by DeepSeek V4 soon. It's capable of reasoning as well as agentic tool use and is essentially a GPT-5, Claude 4.5 Sonnet, and Gemini 3 competitor with the benchmark performance to match. (At least when it comes to tasks involving language; one key feature DeepSeek V3 lacks is multimodality, so it can't handle images or video.)
DeepSeek R1 is a reasoning model. It shocked the AI world when it launched in early 2025. At the time, it was the first non-OpenAI reasoning model to be released and it offered similar performance to o1 and o3-mini despite being freely available as an open model. DeepSeek R1 is no longer cutting edge, and reasoning is integrated with every major frontier model, but it still stands out for the impact it had.
As well as DeepSeek V3 and DeepSeek R1, DeepSeek has a number of older models—including DeepSeek V2, DeepSeek Coder V2, and DeepSeek VL—that you can still download and run yourself but aren't widely available.
Why are DeepSeek-R1 and DeepSeek-V3 so impressive?
DeepSeek R1 and DeepSeek V3 were incredibly impressive when they first launched for a couple of key reasons:
The U.S. had banned the export of Nvidia H100 chips to China, in part to try to prevent Chinese tech companies from developing powerful AI models. Instead, DeepSeek was able to use lower spec H800 chips along with clever architecture optimizations to train its models.
DeepSeek claimed that it was able to train V3 for less than $6 million. This number was a bit misleading as it only represented the GPU rental costs associated with the final training run and not any of the prior research, data gathering, salaries, and all the other expenses necessary to develop frontier models and run an AI company. Still, it was significantly cheaper than what U.S. tech companies were claiming they needed to spend.
DeepSeek, an essentially unknown Chinese AI company, matched OpenAI's most impressive model to that point before Google, Anthropic, or any of the other major players.
Most importantly, though, DeepSeek R1 and V3 set the stage for other models to follow. They're no longer unique or impressive, which is a big deal.
Since the "DeepSeek moment," Chinese AI companies like Alibaba, Moonshot AI, Z.ai, Xiaomi, and ByteDance have all released open models that are competitive with proprietary models from OpenAI, Anthropic, and Google. It represents a major shift in the broader AI ecosystem as proprietary models like GPT-5, Claude, and Gemini 3 are no longer miles ahead of the competition.
In particular, Chinese AI models are good enough for many uses and significantly cheaper to run than proprietary models. Some estimates suggest that as much as 30 percent of global AI usage is from Chinese models.
DeepSeek is not without controversy
As an AI model released by a Chinese tech company, DeepSeek has attracted a fair amount of controversy.
As The New York Times reports, the Chinese Communist Party is instituting an "increasingly complex set of rules" that AI companies have to follow so that they can compete on the international stage without disrupting the "the stability of Chinese society and the Communist Party's hold over it."
You can see this in DeepSeek's hosted chatbot. Unsurprisingly, it has some censorship issues. It refuses to talk about Tank Man and considers Taiwan part of China, for example, though I was able to get it to discuss whether or not China is committing genocide against Uyghur Muslims.
Similarly, DeepSeek is explicitly sending data to China. Various governments, including Australia, South Korea, Canada, the Czech Republic, and Texas, have banned DeepSeek from government devices out of privacy concerns.
Then you get into the real geopolitics. DeepSeek has allegedly procured Nvidia chips that are banned for export to China, though both it and Nvidia deny the reports. OpenAI has also accused DeepSeek of training R1 using ChatGPT's outputs without permission—though the irony of this isn't lost on anyone, and it's never been substantiated.
How much all this matters to you really depends. Every AI company trains their chatbots and models to respond in specific ways to avoid controversy. If you aren't planning on using the DeepSeek chatbot to campaign for Tibetan freedom, you're unlikely to run into any issues with DeepSeek's responses. You can also self-host DeepSeek's models and run them on your own hardware, which further minimize any risks involved.
How to try DeepSeek
The simplest way to check out DeepSeek is through the free DeepSeek chatbot app. It's available on the web, and for iOS and Android devices. Or, if you have the technical skills, you can just grab any of its models from Hugging Face.
While DeepSeek's chatbot is more barebones than ChatGPT in many ways, it works. You don't have any bells and whistles, but you can still use incredibly capable AI models.
How to automate DeepSeek with Zapier
Zapier allows you to connect DeepSeek to thousands of other apps so you can use its affordable frontier models in your existing workflows. For example, you can have DeepSeek respond to Telegram messages or act on data in a Notion database.
Get started with one of these pre-made templates, or start from scratch to bring the power of DeepSeek into any app you use.
Create chat completions in DeepSeek from new data source items in Notion
Issue new api requests in DeepSeek for each new message in Telegram
Create chat completions in DeepSeek every day with Schedule by Zapier
Create DeepSeek chat completions from new or updated rows in Google Sheets
Zapier is the most connected AI orchestration platform—integrating with thousands of apps from partners like Google, Salesforce, and Microsoft. Use forms, data tables, and logic to build secure, automated, AI-powered systems for your business-critical workflows across your organization's technology stack. Learn more.
Related reading:
This article was originally published in January 2025. The most recent update was in February 2026.
