DeepSeek AI Claims Are Shaking the World – The Future of AI?
A Chinese artificial intelligence company, DeepSeek AI, made shocking news this week by claiming that its new AI model is better than OpenAI’s while costing just a fraction to build. These bold claims have sparked major debates in the AI community, with some experts impressed and others highly skeptical.
The Big Claim: AI at a Lower Cost
DeepSeek AI announced that training its large language model cost only $5.6 million—a shocking contrast to the billions that companies like OpenAI, Google, and Anthropic spend on developing their models.
This news triggered concerns about how much money big tech firms are investing in AI infrastructure. Investors reacted strongly, and Nvidia, a major chip manufacturer, lost nearly $600 billion in market value in just one day—the biggest one-day loss for any company in U.S. history.
But is DeepSeek’s claim really true? That’s the big question.
People Also Read: Mastering Google Gemini AI: Top Tips and Best Practices
What is DeepSeek AI?
DeepSeek was founded in 2023 by Liang Wenfeng, co-founder of an AI-focused hedge fund called High-Flyer. The company’s main goal is to develop large language models and eventually create artificial general intelligence (AGI)—an AI that can think, learn, and solve problems like a human.
DeepSeek launched R1, its newest reasoning model, last week. Unlike traditional AI models, reasoning models break down a question into smaller parts and explore multiple ways to solve it before generating an answer. This makes them better at handling complex problems.
While the technology behind R1 isn’t entirely new, DeepSeek claims to be the first to use it in a high-performing AI model while cutting down power consumption.
Is DeepSeek AI Really Different from OpenAI?
DeepSeek’s AI has two main models:
- V3 – Its large language model that powers AI applications.
- R1 – Its reasoning model is designed to improve problem-solving.
A key difference is that DeepSeek’s models are open-source. Thus, it means that developers can freely access, modify, and share the code. On the other hand, OpenAI keeps its most powerful models private.
DeepSeek’s V3 model is also smaller than that of OpenAI. It has 671 billion parameters (AI’s learning variables). In contrast, OpenAI’s latest model is estimated to have over a trillion.
Despite being smaller, DeepSeek claims its R1 model performs just as well as OpenAI’s o1 in reasoning tasks. It backs this up with benchmarks from AIME 2024, Codeforces, GPQA Diamond, MATH-500, MMLU, and SWE-bench Verified. Most surprisingly, DeepSeek says it trained V3 for only $5.6 million, while OpenAI and others spend billions.
If true, this could change the AI industry by proving that advanced AI models don’t have to be expensive to develop. However, many experts aren’t convinced yet.
Related Article: What is ChatGPT? Here’s Everything that You Need to Know
Experts Question DeepSeek’s AI Claims
Tech analyst Daniel Newman, CEO of The Futurum Group, sees DeepSeek’s approach as a “massive breakthrough.” However, he also questions the company’s cost claims.
“I think DeepSeek’s method is a huge step forward, but there are still a lot of unanswered questions about its real costs,” he said.
Another expert, Paul Triolio from DGA Group, explains that DeepSeek’s $5.6 million figure only covered one training run, not the total cost of research and development.
“The full cost was probably much higher,” Triolio said. “But even so, it’s likely still lower than what U.S. AI companies spend.”
DeepSeek vs. OpenAI: A Pricing Comparison
Both DeepSeek and OpenAI charge for using their AI models. Here’s how their pricing compares:
- DeepSeek R1 charges $0.55 per million input tokens and $2.19 per million output tokens.
- OpenAI o1 charges $15 per million input tokens and $60 per million output tokens.
- OpenAI GPT-4o mini (a smaller, cheaper model) charges $0.15 per million input tokens.
This means DeepSeek’s AI is far cheaper to use than OpenAI’s, but it remains to be seen whether it delivers the same level of performance.
Did DeepSeek Break U.S. AI Chip Rules?
Another controversial topic is whether DeepSeek used banned U.S. AI chips to develop its model. The U.S. restricts exports of advanced Nvidia AI chips (like the H100) to China. As a result, it will make it harder for Chinese AI firms to compete.
DeepSeek insists it used older Nvidia chips (H800 and A100), which are allowed in China. However, some experts, like Scale AI CEO Alexandr Wang, doubt this claim and suspect DeepSeek may have used banned chips. In addition, DeepSeek denies any rule-breaking, but this debate is far from over.
In conclusion, the low-cost AI model of DeepSeek AI has sparked excitement and skepticism. If its claims hold up, this could reshape the AI industry, proving that cutting-edge AI doesn’t have to be expensive. However, with experts raising questions about the actual costs, performance, and legality of DeepSeek’s approach, only time will tell if it can genuinely rival OpenAI.
Source: DeepSeek’s AI claims have shaken the world — but not everyone’s convinced