There is a particular race to offer the most advanced artificial intelligence model on the market, and not only among different companies in the United States, but also worldwide with the advancement of certain proposals in the Chinese market.
It is surprising that the Chinese laboratory DeepSeek has launched an open version of R1, that is, its most advanced artificial intelligence-based reasoning model, which rivals, and even surpasses in certain aspects, the o1 model of OpenAI’s ChatGPT.
Specifically, R1 of DeepSeek outperforms o1 of OpenAI en AIME, MATH-500 y SWE-bench Verified.
To understand this, AIME uses other models to evaluate performance; then MATH-500 is an amalgamation of reasoning problems; while SWE-bench Verified focuses on different programming tasks.
That is, being a reasoning model, R1 checks itself.
This makes it a model that may take a little longer to respond than other more general models, but at least it does offer more reliable answers in some environments, such as science, mathematics and physics.
The surprising thing is that this DeepSeek R1 model contains 671 billion parametersparameters that correspond to problem-solving skills.
In any case, they have also published other more simplified versions of this model, even small enough to run on any laptop.
Regarding the complete R1 model, it would not work on a normal laptop, since it requires much more powerful hardware, but at least the API is at a fairly competitive price, at least if we compare it with the OpenAI.
We will see how this technological race between the United States and China progresses to offer the most advanced artificial intelligence model on the market.
Get to know how we work in NoticiasVE.
Tags: Artificial intelligence