DeepSeek has released the AI model R1, which surpasses OpenAI's o1

Chinese company DeepSeek has released an open version of its artificial intelligence model R1, which is considered one of the most advanced in the world. R1 demonstrates results comparable to or surpassing OpenAI's o1 model on the AIME, MATH-500, and SWE-bench Verified benchmarks.

The model is available on the Hugging Face platform under the MIT license, making it free to use even for commercial purposes. R1 has 671 billion parameters, but the company has also introduced lighter versions suitable for running on less powerful hardware. The full version is offered via API at a price 90–95% lower than that of competitors.

R1 has attracted significant attention from the technology industry, particularly against the backdrop of U.S. restrictions on exporting advanced technologies to China. DeepSeek has achieved outstanding results through optimization and collaboration, despite lacking access to cutting-edge chips.

R1 has already become a hit among users — its AI assistant tops the free apps chart in the Apple App Store.

The release of R1 coincided with the U.S. administration's proposal to tighten export controls on AI technologies. Meanwhile, analysts note that Chinese companies such as DeepSeek, Alibaba, and Kimi continue to actively catch up with their Western counterparts.

Source: TechCrunch