Startup DeepSeek will share its AI code repository with all developers and researchers starting next week, allowing anyone to download it and write apps based on it or improve on it.

deepseek wsj
DeepSeek will share a repository of the code and data behind its AI models. Photo: WSJ

With this move, the Chinese company further cements its open-source approach to AI development. DeepSeek’s open-source model surprised everyone by outperforming OpenAI and Meta in several benchmarks.

Companies like Meta have made their models public, but DeepSeek goes further, releasing the source code, the data used to create it, and how they develop and manage the code.

DeepSeek’s move could also escalate the race between the U.S. and China to develop advanced AI models. Making the source code freely available could open up wider applications for DeepSeek’s technology, which governments from the U.S. to Australia have raised concerns about its security.

On the X account, DeepSeek wrote : “We are a small team exploring artificial general intelligence (AGI). Starting next week, we will open source 5 repositories, sharing small progress with complete transparency.”

A code and data repository is a digital storage space that organizes and manages the data and resources needed to train, operate, and evaluate AI models.

DeepSeek’s unexpected success has forced larger rivals like Baidu to adopt an open-source approach. But foreign rivals like OpenAI and Anthropic have kept their AI models, repositories, and data secret.

(According to Bloomberg)