DeepSeek R1: The AI Model That Took the World by Surprise

If there’s one thing the AI world knows how to do, it’s keeping us on our toes. DeepSeek R1 is the shiny new large language model that dropped over Christmas and sent ripples through Silicon Valley—and beyond.

‍

It’s not every day you see industry giants caught off guard, but DeepSeek R1 managed to do just that.

‍

‍

What’s the Big Deal About DeepSeek R1?

DeepSeek R1 was trained using 671 billion parameters—a massive number, sure, but what’s even more impressive is how efficiently it was done.

‍

They only needed 2.7 million GPU hours to train the model. To put that in perspective, that’s a fraction of what many other large language models use. In fact, it’s 11 times less than the GPU time required for a similar model like LLaMA.

‍

Training DeepSeek R1 cost just $6 million. That’s far less than the budgets we’re used to seeing in the AI world. What really stands out, though, is its inference cost—essentially, the cost of running the model.

‍

DeepSeek R1 operates at just $0.27 per million tokens, compared to $3 for other leading models. In simpler terms, it’s not just capable; it’s accessible and budget-friendly, which could open up AI to more people and businesses than ever before. This combination of efficiency and cost-effectiveness is what makes it such a standout.

‍

‍

The Vision Behind DeepSeek

Pascal Coppens, our go-to expert on China, shed light on what makes DeepSeek’s story so remarkable. The team behind it isn’t your typical cohort of corporate engineers chasing profit margins.

‍

Instead, he described them as passionate techies, mostly fresh graduates from top Chinese universities like Tsinghua and Beijing University, who simply wanted to prove they’re the best at what they do.

‍

According to Pascal, they see deep tech as an art form rather than just a job, and it’s this mindset that’s reshaping the AI landscape.

‍

Using limited resources and hardware not as advanced as what’s available in the U.S., they built something extraordinary.

‍

Listen to the full episode of our Radar podcast, where Pascal explores the vision behind DeepSeek R1, the future of AI innovation, and what this means for businesses worldwide. [Click here to tune in!]

‍

The new generation of innovators

As we look to the future, DeepSeek R1 stands as a reminder that the race for AI dominance is no longer defined by who has the most money or hardware.

Instead, it’s about who can dream bigger, work smarter, and challenge the status quo. And for now, the world is watching—and learning—from the techies behind DeepSeek R1.

‍

Stay ahead with nexxworks

🤔 Are you looking to go beyond the headlines and dive into meaningful conversations about the next era of AI. Let’s design a custom tour where you’ll connect directly with the experts driving change.

‍