Chinese developers create Chat GPT competitor

Artificial intelligence developer DeepSeek has released the DeepSeek-R1 model, which is positioned as a strong competitor to OpenAI’s recently launched o1 model. The main advantage of DeepSeek-R1 is that it is free to use thanks to hosting the model on the open Hugging Face platform, unlike the paid subscription to o1, where the ChatGPT Pro tariff reaches $200 per month.

DeepSeek claims that its R1 outperforms o1 in a number of tests, including AIME, MATH-500 and SWE-bench Verified, which assess AI skills in logical reasoning, mathematics and programming. The company’s white paper states that the R1 model has 671 billion parameters, allowing it to effectively solve complex tasks.

DeepSeek has also introduced lightweight versions of the model with 1.5 billion and 70 billion parameters. The simplest model can run autonomously on a standard laptop, while the full version requires the computing power of a large center.

Although the basic version of R1 is available for free, paid access is provided to use the full model via API. However, as the company assures, its cost is 90-95% lower than that of OpenAI.

At the same time, experts warn that R1 training was carried out on data filtered by Chinese regulatory authorities. This may lead to differences in the interpretation of some facts, for example, about the status of Taiwan compared to the Western point of view.