An open version of DeepSeek R1 has been released by the Chinese AI lab DeepSeek, named the ‘reasoning model’. The company claims that this new open-version reasoning model is a tough competitor of OpenAI o1 on certain AI benchmarks.
DeepSeek R1 is free open source and can be used commercially without restrictions as it has an MIT license from the AI dev platform.
What is DeepSeek R1?
DeepSeek R1 is an open-source AI reasoning model developed by the Chinese AI company DeepSeek. This AI reasoning model can perform tasks related to logical inference, mathematical problem-solving, and real-time decision-making.
The DeepSeek R1 reasoning model is way better than OpenAI o1, as it gives users more logic, making it easier to understand and, if necessary, challenge its output. It is essential that outcomes get explained, like research or complex decision-making, and this capability gives the reasoning model an edge.
Also read: Nvidia Geforce RTX 5000 series graphics cards are unveiled
Features of DeepSeek R1 reasoning model:
The most attractive and competitive thing about the DeepSeek R1 model is its open-source nature, which makes it available for all developers and researchers without restrictions.
All developers and researchers can explore, modify, and deploy it within certain technical limits, such as resource requirements.
Being a reasoning model, it takes a longer time to arrive at a solution. However, the information is more reliable than that of the non-reasoning model, mainly in domains such as physics, science, and math.
It has been revealed by DeepSeek that the reasoning model contains 671 billion parameters. These parameters are responsible for a model’s problem-solving skills. More parameters, better performance as compared to fewer parameters.
Also, ‘distilled’ versions of R1 have been released by DeepSeek, as 671 million parameters are massive.
This distilled version size ranges from 1.5 billion parameters to 70 billion parameters. These smaller versions can easily run on a laptop. Full R1 requires beefier hardware, but it is available through DeepSeek’s API at prices 90%-95% cheaper than OpenAI’s o1.
DeepSeek R1 is superior to OpenAI o1 on certain AI benchmarks. According to the company, R1 is superior on the benchmarks AIME, MATH-500, and SWE-bench Verified.
AIME employs other models to evaluate a model’s performance, while MATH-500 is a collection of word problems. SWE-bench Verified, meanwhile, focuses on programming tasks.
One drawback of R1 is that, as it is a Chinese model, it has to be benchmarked by China’s internet regulator to ensure that its responses “embody core socialist values.
But as this powerful reasoning model is available for free, without restriction for researchers and developers, everybody is considering it as a “Beginning of new Era”.
