DeepSeek claims that its proposal beats OpenAI’s o1 in some benchmarks

Chinese AI lab DeepSeek has released an open-source version of DeepSeek-R1, which is called a thinker, which it says works well with OpenAI’s. o1 on some AI indicators.

R1 is available from the AI ​​dev platform Hugging Face under the MIT license, meaning it can be used commercially without restrictions. According to DeepSeek, R1 beats o1 in benchmarks AIME, MATH-500, and SWE-bench Verified. AIME uses some variables to evaluate the model’s performance, while MATH-500 is a list of words. SWE-bench Certified, meanwhile, focuses on software applications.

Being a hypothetical example, R1 checks itself, which it helps to avoid some of the pitfalls that often occur in models. Discussion models take longer – often seconds to minutes longer – to arrive at answers compared to a simple non-intelligent model. Interestingly, they tend to be more reliable in areas such as physics, science, and mathematics.

R1 has 671 billion shares, DeepSeek revealed in a technical report. Parameters correspond to the model’s ability to solve problems, and models with more parameters perform better than those with fewer parameters.

671 billion shares is huge, but DeepSeek also released “distilled” versions of R1 ranging in size from 1.5 billion shares to 70 billion shares. Small things can run on a laptop. As for the R1 as a whole, it needs beefier equipment, but that and available through DeepSeek’s API at prices 90%-95% cheaper than OpenAI’s o1.

There is a drop in R1. Being a Chinese model, it depends benchmarking and China’s internet watchdog to ensure that its solutions “incorporate key social values.” R1 does not answer questions about Tiananmen Square, for example, or Taiwan’s independence.

Most of them China’s AI systemincluding some examples to consider, decrease to respond to topics that may arouse the anger of the country’s rulers, such as speculations about Xi Jinping authority.

R1 will arrive after the days of Biden’s administration are over provided angry export regulations and restrictions on AI technology for Chinese businesses. Companies in China have already been banned from buying AI chips, but if the new rules go into effect as written, companies will face tougher tools in semiconductor technology and the types needed to launch advanced AI systems.

In a policy document Last week, OpenAI urged the US government to support the development of US AI, fearing that Chinese models could match or surpass them. In a interview and The Information, OpenAI’s Vice President Chris Lehane named High Flyer Capital Management, the parent company of DeepSeek, as the organization most affected.

Currently, at least three Chinese labs – DeepSeek, Alibaba, and Likewhich features a Chinese unicorn Moonshot AI – they have created models that they say are o1. (Obviously, DeepSeek was the first – it he announced image of R1 at the end of November.) In a post on X, Dean Ball, an AI researcher at George Mason University, said these developments suggest that China’s AI laboratories will continue to be “fast followers.”

“The success of DeepSeek’s discarded models (…) means that thinkers will continue to multiply and can be controlled by local tools,” Ball wrote, “far from any government surveillance.”