DeepSeek AI: The New Contender in AI

In the rapidly evolving landscape of artificial intelligence (AI), a new name has been making headlines over the past two weeks. Its arrival is said to be a turning point that could reshape the dominance of OpenAI and Google.

On Saturday, January 25, 2025, a Chinese-made AI model emerged as a new competitor to existing Large Language Models (LLMs) like ChatGPT and Gemini. DeepSeek, an AI company founded by Liang Wenfeng in 2023, introduced its latest model, DeepSeek-R1.

Endang Wahyu Pamungkas, S.Kom., M.Kom., Ph.D., a researcher at Universitas Muhammadiyah Surakarta (UMS) specializing in Natural Language Processing (NLP), has been closely observing the rise of DeepSeek. With his extensive experience and deep understanding of AI model development, he sees DeepSeek AI as an innovation born out of limitations. But why is that?

“DeepSeek AI is quite interesting. I recently tested it out and conducted a small observation. This latest Chinese AI tool was developed under constraints, yet it managed to provide a more efficient solution due to its cost-effectiveness,” said the UMS Informatics Engineering lecturer, commonly known as Dadang, in a virtual interview on Tuesday (4/2/2025) 

According to Dadang, a critical factor that led to the development of DeepSeek AI was China’s restricted access to advanced Nvidia GPU chips. Since the United States imposed export limitations on Nvidia A100 and H100, Chinese developers have been forced to find alternative solutions to continue building high-quality generative AI. Instead of being an obstacle, these limitations pushed innovation forward.

A report from Stanford University’s 2024 Artificial Intelligence Index states that DeepSeek developers claimed to need only $5.6 million to train their latest AI model. This is significantly cheaper compared to OpenAI’s ChatGPT-4, which reportedly cost $78 million, or Google’s Gemini Ultra, which cost $191 million.

“This means that businesses or companies looking to adopt AI without making a huge investment in computing infrastructure may find DeepSeek AI an attractive option,” added the UMS Center for Social Informatics Studies researcher.

Looking at netizens' reactions on various media platforms such as X and Instagram, many have started questioning whether DeepSeek AI could replace ChatGPT as the leading generative AI model. According to Dadang, the answer depends on its intended use.

“For everyday needs, like casual Q&A or light conversations, ChatGPT may still be preferable choice because its model is more refined in understanding human language,” he said.

However, he noted that DeepSeek AI has advantages in other areas, particularly in flexibility and cost. Since the R1 model is open source, users have greater control over fine-tuning (the process of adapting a pre-trained model for specific tasks or use cases) according to their needs.

“For instance, if a hospital wants an AI model that is more accurate in understanding medical terminology, they can retrain DeepSeek-R1 with their own dataset,” he explained. This is in stark contrast to ChatGPT, which remains a closed model.

Endang further elaborated that ChatGPT is a proprietary model, meaning users can only use it as is without access to modify its functionality. In contrast, DeepSeek-R1 offers more flexibility, allowing it to be customized for specific needs through fine-tuning.

Beyond flexibility, he also highlighted DeepSeek AI’s advantage in operational costs. He pointed out that ChatGPT’s subscription fee is quite expensive, especially for businesses. Meanwhile, DeepSeek AI offers both free and paid options, with a highly affordable subscription fee.

According to Bloomberg Technoz, DeepSeek users only need to pay around USD 0.50 per month (approximately Rp8,116), a stark contrast to ChatGPT’s subscription fee of USD 20 per month (around Rp324,640).

Despite offering various advantages, DeepSeek AI still faces major challenges, particularly in accuracy and bias. According to an audit by NewsGuard, a service that assesses news and information websites, DeepSeek-R1 currently has an accuracy rate of only 17%, with a failure rate of 83%. It also repeats false claims 30% of the time and provides vague or unhelpful responses in 53% of cases.

Dadang emphasized that accuracy and bias issues are not exclusive to DeepSeek AI, they also affect ChatGPT and other AI models. “Every AI model has biases because bias often comes from the data used to train the model,” he explained.

However, what sets DeepSeek AI apart from its competitors is its transparency. Since it is open source, the model allows researchers and developers to actively identify and improve its weaknesses, whether in terms of accuracy or bias mitigation.

Dadang firmly stated that the AI industry is now shifting toward models that are far more efficient and cost-effective. “Now we can see that companies are no longer just looking for the smartest AI,” he said.

He pointed out that major corporations are prioritizing the economic value of AI rather than simply chasing the highest accuracy. The trend suggests that in the future, AI competition will not only focus on intelligence and precision but also on operational efficiency and cost-effectiveness.

"For instance, if ChatGPT-4 has a 98% accuracy rate but costs Rp1 billion, while DeepSeek-R1 has a 90% accuracy rate but only costs IDR 100 million, many companies would likely choose the second option,” Dadang explained.

“With its energy-efficient, flexible, and low-cost approach, models like DeepSeek AI have a strong chance of reshaping the global AI competition,” he added

Reported by Tempo, the emergence of the dark horse DeepSeek has disrupted OpenAI and Google's dominance. Its presence has posed a significant challenge, even affecting the market valuation of its companies.

Despite its low-cost appeal, Dadang reminded us not to overlook the accuracy issues of the R1 model, which still needs improvement. DeepSeek AI’s success in the global market will largely depend on how well the developer community and businesses adopt this model.

So, can DeepSeek AI truly overtake ChatGPT’s dominance? According to Dadang, the answer to this question may only become clearer and clearer within the upcoming years.

Writer: Genis Dwi Gustati

Translator: Farizal Luqman Majid

Editor: Al Habiib Josy Asheva