COLUMBUS, Ohio — There is a so-called “AI race” happening among tech companies here in the U.S. and in countries like China. 

The release of OpenAI’s ChatGPT caused Chinese companies to rush to compete, which is something they have failed to do in the past. That all changed this year when China-based DeepSeek released its chatbot app. 

Researchers say it rivals ChatGPT’s technology, and the company says it costs a lot less to make.

People may be familiar with AI chatbots like Microsoft’s Copilot, Google’s Gemini and OpenAI’s ChatGPT. They’re apps or interfaces that can have human-like conversation using “large language models” or LLMs to answer questions and solve problems. 

These chatbots are powered by semi-conductor chips made by companies like Nvidia and Intel, which are not cheap to make. 

According to estimates from research firm, Semianalysis, running ChatGPT costs OpenAI around $700,000 per day to operate. 

“The cost is quite shocking,” said Huan Sun, an associate professor and researcher at Ohio State University. “I mean, since ChatGPT, we have been seeing a lot of progress in developing large language models, like different versions of GPT models from OpenAI and then cloud models, but a lot of them are so-called, closed source.” 

Closed source means the code and training data of the model are not accessible or modifiable by the public. 

In January, Chinese company DeepSeek emerged with their R1 model. Sun said its functionality is comparable to ChatGPT’s 01 model and its open source, meaning the software is distributed freely and people can modify it for their own use.

“I think that’s why people get very excited about it, and it is a much more open model that can compete with the best closed source models,” Sun said.

When DeepSeek launched its model, it caused a shakeup in the market, as the company says, it costs significantly less to make and operate. 

Sun said she and her team compared it to ChatGPT.

“We observed both models could achieve comparable performance, like around 30% to 40% performance,” she said. “But then the cost is quite shocking, because DeepSeek R1 cost less than 10% of the cost of OpenAI 01.”

Sun believes more companies will opt to make their interfaces open source, which allows researchers like herself to not only look at the answer a chatbot provides but also how it came to that conclusion.

“Those interesting questions cannot be answered by those closed source models,” Sun said. “So I think this being open source is particularly welcomed by the community, and it’s a trend that’s going to be kept going I think.”

Sun said she expects DeepSeek’s interface to lead to more innovation and development in the months and years to come.