Artificial Intelligence from China

New Chinese AI App Tops the Download List

The Chinese artificial intelligence Deepseek has climbed to the top of the Swedish download list in the App Store, and it has a feature that makes it stand out from other AI models.

It is the Chinese AI company DeepSeek that has released its new AI model DeepSeek R1, and it is said to have comparable performance to OpenAI's AI model o1, which was released last month. DeepSeek already gained some attention last year when they released a chatbot based on their LLM DeepSeek V3, an AI model said to be on par with the AI models GPT-4o from OpenAI and Claude 3.5 Sonnet from Anthropic.

One thing the company has been noted for is its low development costs. DeepSeek states, for example, that it "only" cost the equivalent of about 6 million dollars to train LLM DeepSeek V3, something that can be compared to OpenAI's GPT-4 model, which is said to have cost over 100 million dollars to train.

DeepSeek also states that they trained V3 with only 2000 circuits from Nvidia. AI companies that have developed similar models previously have required 16000 or more Nvidia circuits to train their advanced AI models. This has led some analysts and observers to question whether so much computing power is really needed as today's leading AI companies use to develop their models.

If it turns out that DeepSeek has found a more efficient way to develop AI models than the American AI companies, it could potentially disrupt the playing field for future AI companies and the world's AI development.

Deepseek can be tested in the browser, but is also available as an app for iOS and Android.