Delayed. Again. OpenAI holds back Open-Weight AI model launch: Here’s why 

OpenAI has postponed the release of its first open-weight AI model, citing the need for more safety testing and high-risk reviews before public rollout.

OpenAI has once again delayed the release of its first open-weight artificial intelligence model, citing the need for additional safety tests and review of high-risk areas. The model, initially expected to go live this week, now has no official release date.

In a public statement, CEO Sam Altman said the team had planned to launch the model next week but decided to hold back in order to ensure it meets safety standards. “We need time to run additional safety tests and review high-risk areas. We are not yet sure how long it will take us,” Altman stated.

He added that while OpenAI trusts the developer community to build great things, “once weights are out, they can’t be pulled back. This is new for us and we want to get it right.”

What is the Open-Weight model?
An open-weight AI model refers to a machine learning system whose underlying architecture and weights are made publicly available. Developers and researchers can freely download, modify, and deploy these models—unlike closed systems such as GPT-4 or Claude, which remain fully proprietary.

OpenAI has long been under pressure to release such a model, especially after rival companies released successful open-weight alternatives. The growing popularity of open AI systems among enterprises and independent developers has added to this demand, primarily because of the control, customisation, and transparency these models offer.

Why the delay?
OpenAI’s leadership has emphasised that the release of such a model represents a major shift for the company. The irreversible nature of publishing model weights means the potential risks—such as misuse or unintended capabilities—must be thoroughly evaluated.

Internally, OpenAI teams have been running extensive capability and safety tests. While the model is reportedly ready from a performance perspective, concerns remain about how it might be used or adapted once it’s publicly accessible. The company has also acknowledged that this is a new experience for them and that they prefer to “get it right” rather than release prematurely.

This isn’t the first delay. The model was initially expected to debut last month, but OpenAI had pushed back the release due to internal changes in direction. At the time, Altman hinted that the research team had developed something “unexpected and quite amazing,” but that it required more time to be polished and made safe.

The model in question is expected to feature advanced reasoning capabilities and serve as an answer to the growing dominance of other open-weight models in the market.

The repeated delay from OpenAI highlights the growing tension between innovation and safety in the generative AI race. While open-source and open-weight models are rapidly evolving and becoming more powerful, releasing them too quickly could pose risks—from disinformation to unauthorised commercial use.

Meanwhile, AI labs across the globe are moving ahead with open-weight strategies, gaining developer mindshare and enterprise traction. OpenAI’s cautious approach may slow momentum but reflects a commitment to responsible AI development.