OpenAI releases open-weight models after DeepSeek’s success

OpenAI CEO Sam Altman previously said his company needs to “figure out a different open source strategy.”

OpenAI is releasing a pair of open and freely available artificial intelligence models that can mimic the human process of reasoning, months after China’s DeepSeek gained global attention with its own open AI software.

The two models, called GPT-oss-120b and GPT-oss-20b, will be available on AI software hosting platform Hugging Face and can produce text — but not images or videos — in response to user prompts, OpenAI said on Tuesday. These models can also carry out complex tasks like writing code and looking up information online on a user’s behalf, the company said.

Crucially, the models are both open-weight systems, similar to Meta Platforms Inc.’s Llama. The term “weight” refers to the parameters in an AI model. OpenAI is disclosing the many numerical values the models picked up and were tweaked with during the training process, allowing developers to better customize them. However, OpenAI is not revealing the data used to train them, falling short of the definition for a truly open source AI model.

Despite its name, most of OpenAI’s models are closed systems — the kind of software that’s controlled by the developer, can’t be modified by users, and includes less transparency about its technical underpinnings. Like many of its US rivals, OpenAI has guarded its training data and focused on charging more for its most powerful models to offset the immense cost of development. OpenAI has not released an open model since 2019 when it rolled out GPT-2, a predecessor to the software that originally powered chatbot ChatGPT.

Shortly after DeepSeek released its open R1 system in January, OpenAI Chief Executive Officer Sam Altman said the company was discussing releasing some model weights. He also acknowledged OpenAI needed to “figure out a different open source strategy.” More recently, President Donald Trump’s AI Action Plan encouraged the development of open models, citing their potential to “become global standards” in business and academic research.

However, critics of open source software have long argued that it’s less secure. In July, Altman posted on social network X that the company had planned to launch an open-weight model in the middle of that month, but was holding off “to run additional safety tests and review high-risk areas.”

“While we trust the community will build great things with this model, once weights are out, they can’t be pulled back,” Altman said at the time. He added that such a release is “new for us” and the company wanted to “get it right.”

OpenAI intends for the new open models to be used by individuals, companies and governments that want to tweak and run AI systems on their own hardware and services. Orange SA and software company Snowflake Inc. are among the companies that have been testing out the models, OpenAI said.

Both new models are relatively compact and efficient. The more powerful option, 120b, can run on one 80 gigabyte graphics processing unit, the company said. The 20b model, meanwhile, can operate on a laptop with 16 gigabytes of memory, OpenAI said.

During a briefing with reporters this week, OpenAI said it hopes to receive feedback from users that it can then review to decide what kind of open model to release in the future.