In a twist so shocking it almost feels predictable, OpenAI CEO Sam Altman announced Friday that the company is once again delaying the release of its much-hyped open model — this time indefinitely — citing the need for more “safety testing.”
The model was originally scheduled to drop earlier this summer, then postponed to next week, and now it’s been yeeted into the future with no ETA in sight.
“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 posted on X, formerly Twitter, as if to reassure developers that everything is fine, just fine.
“While we trust the community will build great things with this model, once weights are out, they can’t be pulled back,” he added, in what might be the understatement of the year.
The open model is supposed to be OpenAI’s grand re-entry into the free-range AI ecosystem — a rare gesture from a company that has spent the last few years locking its best tech behind increasingly monetized APIs and partner deals.
Unlike the upcoming GPT-5, which is still wrapped in secrecy and likely to be subscription-only, the open model was expected to be downloadable and runnable on local machines.
But now, that promise is trapped in limbo.
OpenAI insists the delay is all in the name of responsibility — because nothing says “we totally have this under control” like postponing a product multiple times after hyping it as the summer’s blockbuster AI release.
Aidan Clark, OpenAI’s vice president of research leading the open model team, echoed the vague sentiment in a post of his own on Friday.
“Capability-wise, we think the model is phenomenal — but our bar for an open source model is high and we think we need some more time to make sure we’re releasing a model we’re proud of along every axis,” he said, without explaining what exactly those axes are or what OpenAI is even afraid of.
This isn’t the first time OpenAI has played coy with breakthroughs.
Back in June, Altman vaguely teased that the company had achieved something “unexpected and quite amazing” with the model — without saying what that was.
Apparently, “amazing” now means “not ready for public consumption.”
Meanwhile, competitors aren’t slowing down to overthink it.
Earlier on Friday, Chinese AI startup Moonshot AI dropped Kimi K2, a one-trillion-parameter model that not only calls itself open but also reportedly outperforms OpenAI’s GPT-4.1 on several advanced coding benchmarks.
So while OpenAI dithers over what constitutes a “safe” model, others are shipping — and, crucially, releasing — actual products.
But now, with delays stacking and rivals flexing, that title seems increasingly at risk.
There have also been internal discussions about allowing the open model to interface with OpenAI’s cloud models for more complex tasks.
Yet even those speculative features are now in question.
For developers, researchers, and tinkerers eager to get their hands on a truly open model from the company that literally has “Open” in its name, the message is clear: wait longer, wait quietly, and definitely don’t ask too many questions.
Because in the world of frontier AI, where transparency was once the mission, delays and vague statements are the new normal — and “open” is starting to sound like a brand, not a principle.
Leave a Reply