OpenAI indefinitely postpones release of open models 'to further verify their safety'



OpenAI CEO Sam Altman announced that OpenAI will postpone the release of its open model, which was scheduled for July 2025. The release of this open model has already been postponed once, but Altman said, 'We will postpone the release indefinitely to further verify its safety.'

OpenAI delays the release of its open model, again | TechCrunch
https://techcrunch.com/2025/07/11/openai-delays-the-release-of-its-open-model-again/

The release of OpenAI's open model was much anticipated, along with the release of GPT-5. Unlike the closed model GPT-5, the open model was freely downloadable and run by developers locally, so there was a lot of anticipation.

According to TechCrunch, an IT news site, OpenAI's open model is considering a very permissive license with almost no restrictions on personal or commercial use. In addition, OpenAI aims to be at the top of benchmark scores compared to other open models such as DeepSeek, Llama, and Gemma, and CEO Altman also said, 'We have achieved unexpected and surprising results. We think it is well worth the wait, but it will take a little more time,' and attention is also being paid to its performance.



Altman announced on X (formerly Twitter) in April 2025 that the open model would be released in the summer of 2025. However, in June 2025, he announced that the release would be postponed, saying, 'We're going to take a little more time with the open weight model. That means we're planning on releasing it later this summer instead of in June.'



Then, on July 12, 2025, CEO Altman commented, 'The reason for the delay is that we need time to conduct additional safety testing and identify high-risk areas. We don't know yet how long that will take. We trust the community will build great things with this model, but once we put the weighting in, there's no going back. This is new for us and we want to do it right. Sorry to be the bearer of bad news. We're working hard,' announcing the indefinite postponement of the release.



Following Altman's comments, Aidan Clark, OpenAI's vice president of research, said, 'We think it's a great model in terms of functionality, but our standards for open source models are high, and we need a little more time to ensure we have a model that we can release with confidence in all aspects.'



As for why Altman postponed the release just before it was released, social news sites HackerNews and X have pointed out that it may have been because of the announcement of Moonshot's open source AI model ' Kimi K2 '. Kimi K2 has been touted for outperforming models such as GPT-4.1 and DeepSeek-V3 in major benchmarks, and it is said that OpenAI feared that the 'unexpected and surprising results' of the open model would be overshadowed by the appearance of Kimi K2 just before the release.

Chinese AI model 'Kimi K2' that exceeds GPT-4.1 is released for free, beating GPT-4.1 and Claude 4 Opus in multiple tests and also handling agent tasks - GIGAZINE

Related Posts:

in Software, Posted by log1i_yk