DeepSeek recommends these settings when deploying its R1 model

It wasn’t a while ago when Chinese-based startup DeepSeek shocked the AI world with its R1 model and an AI chatbot that’s more powerful than ChatGPT. The company now shares its key recommendations if you want to deploy the model.

For optimal results, DeepSeek suggests not using a system prompt so that the model can adapt more flexibly to different inputs. And, you can also the temperature to 0.6 for a good balance between creativity and coherence in the model’s responses.

DeepSeek also offers official prompts for search and file uploads, which users can easily access through a link. To make sure the model works properly and doesn’t produce unwanted results, DeepSeek has created guidelines to help avoid any issues.

The official DeepSeek-R1 deployment runs on the same model as the open-source version, meaning you will experience its full capabilities across various apps.

Though, there are still some concerns over privacy and security in the US, China’s number-one arc rival in the AI boom.

Analyst Charles Archer predicts three potential explanations for the success: one, reverse-engineering existing open-source AI; two, the Chinese Communist Party-backed funding for these older Nvidia GPUs; or three, how the Chinese AI company innovatively scaled with older hardware.

Now, while all those scenarios may actually happen, especially considering China’s past of overly exaggerating its tech advances, the third is what everyone is excited about and feared at the same time.

So, in the States, companies that integrate the model (including Perplexity, the AI search engine, and Microsoft on its Copilot+ PCs), needed to make some tweaks to make the model run locally or host in the US.

The Redmond company has also integrated DeepSeek-R1 into its Azure cloud platform and GitHub tools.