DeepSeek-R1 Can Combine Real-Time Data With Reasoning, Beating OpenAI o1 In Some Benchmarks

DeepSeek

China’s vast size and heavily shielded market from foreign influences offer a significant advantage to domestic companies.

With strict censorship and policies that restrict non-approved products, the country is off limits to many Western innovations. But knowing that Large Language Models are the hype of technology, thanks to OpenAI’s launch of ChatGPT, China’s tech sector has also risen to the challenge.

Numerous companies, both large and small, are entering the AI arena to compete in this rapidly growing and profitable space.

One such contender is DeepSeek.

Its flagship product, DeepSeek-R1, has drawn significant attention for its ability to compete directly with many advanced AI models from the West.

And remarkably, the DeepSeek-R1 is able to rival OpenAI o1 on certain AI benchmarks.

According to DeepSeek, the R1 beats o1 AIME, MATH-500, and SWE-bench Verified.

This innovation underscores China’s growing strength in the AI industry and highlights how its unique market dynamics foster the rise of competitive alternatives to Western technologies.

This is possible because being a reasoning model, like the o1, the R1 effectively fact-checks itself, which helps it to avoid some of the pitfalls that normally trip up models.

And just like the o1, the R1 as a reasoning model takes a little longer — usually seconds to minutes longer — to arrive at solutions compared to a typical nonreasoning model.

The upside is that they tend to be more reliable in domains such as physics, science, and math.

And one of the things that it's capable of, is simultaneous web browsing and reasoning capabilities.

This enhancement allows the AI to access real-time web data while leveraging its advanced reasoning framework, thereby improving its ability to provide accurate and up-to-date responses.

Users can access this feature via the DeepSeek API or its web-based chat platform by activating the "Deep Thinking" mode.

The integration mirrors functionalities seen in models like Perplexity AI, which specialize in combining reasoning with live data retrieval.

R1 contains 671 billion parameters, DeepSeek revealed in a technical report. Parameters roughly correspond to a model’s problem-solving skills, and models with more parameters generally perform better than those with fewer parameters.

Indeed, 671 billion parameters is massive, but DeepSeek also released “distilled” versions of R1 ranging in size from 1.5 billion parameters to 70 billion parameters. The smallest can run on a laptop. As for the full R1, it requires beefier hardware, but it is available through DeepSeek’s API at prices 90%-95% cheaper than OpenAI’s o1.

And unlike the o1, DeepSeek-R1 is available from the AI dev platform Hugging Face under an MIT license, meaning it can be used commercially without restrictions.

Trained with 671 billion parameters, DeepSeek-R1 is certainly massive. .

But for those who wish for a slimmer version, DeepSeek has "distilled" versions of R1 ranging in size from 1.5 billion parameters to 70 billion parameters.

The smallest can run on a laptop.

Whereas the the full capacity of R1 requires powerful hardware, and that users can also pay to use it through DeepSeek’s API at prices 90%-95% cheaper than OpenAI’s o1, the smallest version can run on just a laptop.

While DeepSeek-R1 is a powerful and a worthy rival to OpenAI's products, there is one major downside to it: it's from China.

Being an AI model from China, DeepSeek had to train the R1 using data that has been regulated by China’s internet regulator to ensure that its responses "embody core socialist values."

As a result of this, the R1 will avoid answering questions about Tiananmen Square or Taiwan’s autonomy, for example.

About a week before this, DeepSeek released a dedicated app for its AI for iOS and Android.

Powered by world-class DeepSeek-V3. the app is free to use, cross-platform history and sync, web search and DeepThink, file upload and text extraction.

DeepSeek also said that the app has no ads and no in-app purchases.

Published: 
20/01/2025