format(webp))
DeepSeek asserts its own "reasoning" model is better than OpenAI's o1 in certain tests.
The launch of DeepSeek-R1, a reasoning model from Chinese AI lab DeepSeek, which is now available on the AI development platform Hugging Face. R1 is notable for its large parameter count and ability to fact-check its outputs, enhancing its reliability in complex problem-solving scenarios. While it has been well received, with numerous derivative models created by developers, the model's responses are regulated by Chinese authorities, restricting discussions on sensitive political topics. Overall, DeepSeek-R1 represents a significant advancement in AI reasoning capabilities while navigating the complexities of regulatory compliance.
Chinese AI lab DeepSeek has launched an open version of its reasoning model, DeepSeek-R1, which it claims performs comparably to OpenAI’s o1 on certain AI benchmarks.
R1 is available on the AI development platform Hugging Face under an MIT license, allowing for unrestricted commercial use. According to DeepSeek, R1 outperforms o1 on benchmarks such as AIME, MATH-500, and SWE-bench Verified. AIME uses other models to assess performance; MATH-500 consists of a series of word problems; and SWE-bench Verified focuses on programming tasks.
As a reasoning model, R1 can fact-check its own outputs, helping it avoid common errors that other models might encounter. While reasoning models typically take longer—ranging from seconds to minutes—to reach solutions compared to standard models, they tend to be more reliable in fields like physics, science, and mathematics.
DeepSeek disclosed in a technical report that R1 has 671 billion parameters, which are indicative of a model's problem-solving capabilities; generally, models with more parameters perform better. While 671 billion parameters is substantial, DeepSeek has also released "distilled" versions of R1, with sizes ranging from 1.5 billion to 70 billion parameters. The smallest version can run on a laptop, while the full R1 requires more powerful hardware but is accessible through DeepSeek’s API at prices 90%-95% lower than OpenAI’s o1.
Clem Delangue, CEO of Hugging Face, noted in a post on X that developers on the platform have created over 500 "derivative" models of R1, which have collectively been downloaded 2.5 million times—five times more than the official R1 model.
However, there is a drawback to R1. As a Chinese model, it is subject to oversight by China’s internet regulator, which ensures that its responses align with "core socialist values." Consequently, R1 will not respond to questions regarding topics like Tiananmen Square or Taiwan's autonomy.
What's Your Reaction?
