What is DeepSeek AI?

A new open-source competitor to ChatGPT

DeepSeek, a Chinese startup headquartered in Hangzhou, has garnered significant attention worldwide in recent weeks. Yesterday (January 27, 2025), the model caused the stock market to lose 6000 billion SEK in one day. The company offers AI models that perform as well as or better than industry-leading alternatives from the US – but at a fraction of the cost. This has not only made DeepSeek a global talking point but also a competitor that major AI players now must take seriously.

Affordable innovation challenging the giants

DeepSeek has just launched two models, DeepSeek-V3 and DeepSeek-R1, which are said to be on par with the most advanced models from OpenAI, Meta, and Anthropic. Additionally, DeepSeek has managed to keep its costs low: the V3 model was trained for under 6 million dollars, an impressive feat compared to what American companies typically spend on similar projects. The R1 model, which is equivalent to ChatGPT's o1, is 20 to 50 times cheaper to use. For businesses and developers, this means access to powerful AI technology without breaking the budget.

DeepSeek tops the App Store

One of the company's products, DeepSeek AI Assistant, has quickly become the highest-ranked free app in Apple's App Store in the US, surpassing OpenAI's ChatGPT. The same has happened in Sweden, where DeepSeek tops the list under free apps.

Skepticism and political dimensions

Despite its successes, DeepSeek has also faced skepticism. Critics point out that the company's claimed training costs may be underestimated. Additionally, rumors have circulated that DeepSeek uses advanced Nvidia chips that are banned from export to China, raising questions about how the company actually obtained its hardware.

Chinese authorities view DeepSeek as an important asset in their pursuit of self-sufficiency in strategic sectors like AI. Founder Liang Wenfeng has even participated in meetings with China's Premier, suggesting that the company plays a crucial role in China's technological future plans.

What happened next? (updated 2026)

A year on, it's clear DeepSeek wasn't a one-off. The company has kept shipping models at a rapid pace — and, above all, openly. Every model is released under an MIT license, meaning anyone can download, fine-tune and run them themselves.

• V3 was refined (V3-0324) with better reasoning, partly using techniques borrowed from R1.
• In April 2026, DeepSeek-V4 (V4-Pro and V4-Flash) arrived with clearly stronger agent and reasoning capabilities.
• The much-anticipated R2, however, has been held back. Reports say founder Liang Wenfeng kept it in the drawer because he wasn't satisfied with its performance — an unusual move in an industry that otherwise races to ship first.

The point still stands: open, cost-effective models put pressure on the whole market. That's good news for anyone who wants to build with AI without being locked into a single vendor.

What can we learn from DeepSeek?

DeepSeek is a clear example of how rapidly the AI industry is evolving and how geographical barriers are beginning to blur. It also reminds us of the importance of cost-effectiveness and innovation in creating competitive solutions.

For businesses, this is a reminder that it's not just about big budgets – the right strategy and tools can be decisive.

Want to take the next step in understanding and using AI in your business? Techlära offer consulting services to help you stay ahead.

Keep an eye on DeepSeek, they're just getting started.