The R1 model released by Chinese company DeepSeek has impressed those in the industry with all the features it shows.

Similar to OpenAI o1, the R1 model released by DeepSeek on Jan 20 achieves performance comparable at 3 percent of the cost. Deep-Seek R1 is a first-generation reasoning model, trained via large-scale reinforcement learning, like OpenAI o1.

However, unlike OpenAI o1, Deep-Seek R1 is totally open-sourced, allowing developers and professionals and everybody easy access to the codes it has used in developing the model.

Just like DeepSeek said on github.com, "The open source Deep-Seek-R1, as well as its API, will benefit the research community to distill better smaller models in the future".

Compared with the existing giants, DeepSeek, founded in 2023, is a young company. The small model that DeepSeek has chosen to develop is also a practical strategy for both itself and young enterprises like it in the industry. Currently the mainstream AI models describe things with parameters. For example they describe a banana with parameters that include "yellow","strip-shaped", "edible" and "sweet".The number of parameters is an essential measurement of AI models, as the model that can accurately describe an object with fewer parameters has higher efficiency.

DeepSeek V3, one of the opensource models released by Deep-Seek, which shows higher efficiency, had only 671 billion parameters, of which 37 billion were activated in usage, both of which are quite low in the industry. More important, through deep exploration, it managed to train that model using only 2048 H800 GPUs of Nvidia at the cost of $5.6 million, which is just a small percentage of the cost incurred by OpenAI and Google for training similar types of models.

As the US, guided by its protectionist strategy, is still blocking exports of high-performance chips from its own and its allies' producers to China, the low cost and easy-to-obtain chips are essential for other Chinese developers. For young, innovative Chinese companies to grow into capable players on the global stage, that's the correct direction.