When DeepSeek released its DeepSeek reasoning model in early 2025, it tanked the US stock market. The Chinese startup showed a new way of training frontier AI that didn’t rely on access to expensive, high-end hardware. DeepSeek used software to make up for the premium Nvidia chips it couldn’t buy.
That’s not the only thing DeepSeek did to induce panic. The startup released DeepSeek as an open-source model, allowing anyone on Earth to install the AI on their computers and run it locally. That challenged US AI firms, which didn’t necessarily have open-source alternatives to DeepSeek.
Suddenly, you didn’t have to pay for a premium AI experience via ChatGPT, Gemini, or something else. You had an alternative that was as good as the best ChatGPT model. I’ll point out that other open large language models existed before DeepSeek and still exist. Meta is a good example of that.
Despite the company name, OpenAI, ChatGPT wasn’t available as an open-weight model. Soon after DeepSeek arrived, Sam Altman & Co. announced they would release an open ChatGPT version this summer. But on Friday, Altman had to announce a second delay for the open ChatGPT version, saying the company needs more time to test for safety.
“We are going to take a little more time with our open-weights model, i.e. expect it later this summer but not June,” Altman tweeted on June 11. “Our research team did something unexpected and quite amazing, and we think it will be very, very worth the wait, but needs a bit longer.”
Altman never explained the “unexpected and quite amazing” teaser for the open ChatGPT version, and one shouldn’t get too excited about marketing speak on X from the leader of one of the most important AI companies in the world. Of course, the open ChatGPT version should be amazing. There’s no point in releasing it otherwise.
Also, I’m certain OpenAI would want to give the open ChatGPT version its own flair, so it can attract scientists and developers.
Altman returned on July 12 with an update on the open ChatGPT, announcing another round of delays. Reports preceding his new tweet said OpenAI was about to release the open-weight model, and he confirmed those rumors.
“We planned to launch our open-weight model next week,” Altman said on Friday. “We are delaying it; we need time to run additional safety tests and review high-risk areas. We are not yet sure how long it will take us.”
“While we trust the community will build great things with this model, once weights are out, they can’t be pulled back. This is new for us and we want to get it right,” the CEO added. “Sorry to be the bearer of bad news; we are working super hard!”
How powerful will the open ChatGPT be?
Citing security is a good reason to delay the release of any high-end AI model, whether it’s closed or open. Safety is especially important for the latter. Say OpenAI launches a closed version of ChatGPT that has certain safety issues. In that scenario, OpenAI can still pull the model and patch everything quickly so bad actors in the wild can’t take advantage of certain capabilities.
Remember that models like ChatGPT o3 and similarly advanced alternatives know how to make bioweapons. It’s their training that prevents them from helping malicious actors. That’s just one type of nefarious activity that can benefit from AI use.
The open ChatGPT model that can be installed locally on computers might be as powerful as ChatGPT o3-mini. Any safety issues discovered after launch won’t be patchable universally. For example, a malicious actor would have to install that version of the open ChatGPT that can be jailbroken on their computer. There won’t be any automatic software updates to be performed after that.
That’s just speculation at this point. It’s unclear what sort of features the open version of ChatGPT will offer or what kind of hardware it’ll support.
That said, OpenAI can’t afford not to have an open ChatGPT version out in the wild, not as long as other rivals keep making such AI software available to users.
