OpenAI pushes its open-weight model launch to late summer after discovery of a surprising” breakthrough, prioritizing safety and performance
Altman stated that OpenAI had intended to release the model next week; however, the company is delaying its release indefinitely to conduct additional safety testing.
“We require additional time to conduct safety tests and evaluate high-risk areas.” We are uncertain as to the duration of the process,” Altman stated in a post on X.
While we are confident that the community will achieve amazing things with this model, retracting weights once they have been removed is impossible. It is our first experience with this, and we are eager to ensure it is executed accurately.
The release of OpenAI’s open model is one of the most eagerly anticipated AI events of the summer, in addition to the expected release of GPT-5 by the ChatGPT creator. In contrast to GPT-5, OpenAI’s open model will be accessible to developers for free download and local execution.
Through both launches, OpenAI will endeavor to prove that it remains the premier AI center in Silicon Valley. This is becoming more challenging as xAI, Google DeepMind, and Anthropic invest billions of dollars in their respective endeavors.
Due to the delay, developers will have to wait an additional period to experiment with the first open model that OpenAI has released in several years.
TechCrunch previously reported that OpenAI’s open model is anticipated to possess reasoning capabilities comparable to the company’s o-series of models and that OpenAI intended for it to be the best-in-class among other open models.
This week, the ecosystem of open AI models experienced a slight increase in competition. Moonshot AI, a Chinese AI startup, introduced Kimi K2, a one-trillion-parameter open AI model that outperforms OpenAI’s GPT-4.1 AI model on numerous agentic-coding benchmarks earlier on Friday.
Altman acknowledged that OpenAI had accomplished something “unexpected and quite amazing” when he announced the initial delays around the open model in June. However, he did not provide further details.
In a post on X Friday, Aidan Clark, OpenAI’s VP of research, who leads the open model team, stated, “We believe the model is phenomenal in capability.
However, our standards for an open source model are high, and we believe that we require additional time to ensure that we are releasing a model that we are proud of on all fronts.”
