OpenAI has announced another delay in the launch of its much-awaited open AI model, emphasizing the need for extensive safety evaluations.

This follows a previous postponement from the initial summer release timeline, marking the second delay in recent months.

Sam Altman, CEO of OpenAI, shared the update via a post on X on Friday, stating that the model’s release, originally slated for next week, has been deferred. “Additional safety assessments and a thorough review of high-risk components are necessary,” Altman explained. “Releasing model weights is irreversible, so we’re taking extra precautions to ensure we get this right, as it’s our first open model in years.”

The forthcoming model was set to be OpenAI’s first fully accessible AI system in years, allowing developers to download and operate it locally—a notable shift from the company’s traditionally closed-source strategy.

The announcement coincides with intensified competition in the open-source AI sector. On the same day, Chinese startup Moonshot AI unveiled its Kimi K2 model, boasting one trillion parameters and reportedly surpassing OpenAI’s GPT-4.1 in key coding benchmarks.

Aidan Clark, OpenAI’s Vice President of Research leading the open model project, also addressed the delay on X. “The model’s capabilities are impressive, but our standards for an open-source release are rigorous. We need more time to ensure it meets our expectations across all dimensions,” Clark noted.

Industry expectations had pegged the open model as comparable to OpenAI’s o-series models in reasoning prowess, with reports from TechCrunch suggesting the company aimed to set a new standard for open-source AI systems.

While OpenAI’s reputation in AI innovation remains strong, the delay highlights the complexities of balancing cutting-edge development with safety and ethical considerations. The AI landscape is growing increasingly competitive, with players like xAI, Google DeepMind, and Anthropic accelerating their efforts in advanced model development.

In June, Altman teased the model’s potential during the first delay announcement, describing the team’s progress as “remarkable and unexpected,” though he provided no further details.