The planned release date for OpenAI’s first open-centric language model has been postponed. Company CEO Sam Altman stated that the model was expected to be released next week, but that further security testing and high-risk areas needed to be assessed. The development was announced publicly in a post from Altman’s personal X account.

OpenAI is officially postponing its open AI model

In his post, Altman stated that the model’s release was planned but has been postponed, saying, “We need time to conduct additional security testing and review high-risk areas.” No timeline was given for the new release date. Altman said, “We’re not sure yet how long it will take.”

In the same post, Altman stated that he believes the community will do “great things” with this model, but emphasized that releasing the model is an irreversible step. He emphasized that this is a new experience for the company and that the process must be carried out correctly. The statement concluded, “I hate to be the bearer of bad news, but we’re working very hard.”

OpenAI’s announcement comes as the company prepares to launch its first publicly available open-source model, following the success of its closed-source models. Previous information suggested that the postponed model would have advanced reasoning capabilities and features similar to the o3 mini model, currently available to a limited number of users.

This model would be OpenAI’s first open-source model since GPT-2, released in 2019. However, the term “open-source” does not necessarily mean that the model will be fully open-source.

OpenAI plans to make its training data public, but users will not be allowed to directly modify the model. This suggests the model will offer a degree of transparency, but will be limited in nature to prevent uncontrolled distribution.

The postponement decision comes at a time when security priorities in AI are at the forefront. In recent months, concerns have grown among the public and expert circles about the use of AI in high-risk areas. OpenAI’s unprecedented security sensitivity prior to the release of this model is evaluated within this framework.