DeepSeek Promises to Share Even More AI Code in a Rare Step

Article content

(Bloomberg) — Chinese AI sensation DeepSeek plans to release key codes and data to the public starting next week, an unusual step to share more of its core technology than rivals such as OpenAI have done.

Article content

Article content

The 20-month-old startup, which surprised Silicon Valley with the sophistication of its AI models last month, plans to make its code repositories available to all developers and researchers. That allows anyone to download and build on or improve the code behind the well-regarded R1 or other platforms, it said in a post on X.

Article content

With the move, DeepSeek is pushing harder on an open-source approach to AI development that’s won more advocates since its models outperformed OpenAI and Meta Platforms Inc. competitors in benchmark tests. Companies such as Meta already make their models available to the public, allowing users to customize the platform for their own applications. OpenAI began as partially open source, though it’s since retreated from that mission. But DeepSeek says it intends to go further by publicizing the underlying code, the data used to create it, and the way it develops and manages that code. 

It also potentially escalates a race between the US and China to develop ever-more advanced AI models. By making its coding secrets freely available, DeepSeek is helping to ensure wider adoption of its technology, which is already spurring concerns about security among governments from the US to Australia. 

“We’re a tiny team exploring AGI. Starting next week, we’ll be open-sourcing 5 repos, sharing our small but sincere progress with full transparency,” DeepSeek announced on its X handle on Friday.

Article content

A code and data repository is a digital storage space where the data and resources needed for training, running, and evaluating AI models are organized and managed. The Hangzhou-based startup said its technology has been fully tested, deployed and documented. 

DeepSeek’s surprising progress has forced larger, more established rivals like Baidu Inc. to adopt the open-source framework. But global competitors like OpenAI and Anthropic still keep their AI models, repositories and data proprietary.

Investors in the biggest US AI startups like Anthropic PBC and xAI have plowed tens of billions of dollars into the industry in the hope of a big payday. DeepSeek, which emerged out of a quantitative hedge fund run by founder Liang Wenfeng, has so far not revealed outside backing and could face less pressure to build a revenue model.

“No ivory towers – just pure garage-energy and community-driven innovation,” the startup posted on X.

Article content