Sam Altman: OpenAI will convey again GPT-4o after person backlash

After a GPT-5 rollout that was “just a little extra bumpy” than anticipated, OpenAI CEO Sam Altman mentioned the corporate is bringing again GPT-4o — with a catch.

Throughout a Reddit AMA (and in addition on X), Altman mentioned GPT-4o will quickly be obtainable once more for paid ChatGPT customers.

“Okay, we hear you all on 4o,” mentioned Altman. “Thanks for the time to present us the suggestions (and the eagerness!).” And on X, Altman confirmed, “We’ll let Plus customers select to proceed to make use of 4o. We’ll watch utilization as we take into consideration how lengthy to supply legacy fashions for.”

So GPT-4o lives to combat one other day.

This Tweet is presently unavailable. It may be loading or has been eliminated.

Why is OpenAI bringing again GPT-4o?

OpenAI launched the extremely anticipated GPT-5 mannequin on Thursday, promising the neatest, quickest mannequin but. GPT-5 additionally has a streamlined system that works behind the scenes to decide on which model of GPT-5 is greatest for the person’s immediate. Beforehand, customers may choose particular fashions for every of their prompts. However quickly after the rollout started, ChatGPT customers roasted GPT-5 on Reddit and different social media platforms, calling it a “catastrophe” and “largest piece of rubbish.”

A lot of the criticism got here from GPT-5’s colder “character,” in comparison with GPT-4o. Customers additionally complained in regards to the abrupt removing of all different fashions from the ChatGPT mannequin picker in favor of the ostensibly unified GPT-5. Mockingly, customers beforehand criticized OpenAI for the complicated variety of mannequin choices, proving you could’t please everyone.

Sam Altman and different OpenAI management had a pre-planned Reddit AMA on Friday. What they encountered was a flood of demand to convey again GPT-4o.

Within the AMA, Altman additionally promised ChatGPT customers that GPT-5 will quickly enhance.

“GPT-5 will appear smarter beginning at the moment,” Altman mentioned. “Yesterday, the autoswitcher broke and was out of fee for a bit of the day, and the outcome was GPT-5 appeared approach dumber. Additionally, we’re making some interventions to how the choice boundary works that ought to make it easier to get the correct mannequin extra typically.”

Many customers mentioned they preferred the heat and empathy from GPT-4o (and GPT-4.5), with some evaluating the scenario to shedding a pal in a single day. Nonetheless, GPT-4o additionally had main issues with sycophancy, prompting OpenAI to apologize and tweak the mannequin. Nonetheless, some customers grew very connected.

Some customers complained they did not get advance discover about earlier fashions being faraway from ChatGPT. Nonetheless, it was implied that GPT-5 was developed to exchange them.

“A giant a part of our purpose right here is to get the brand new system to a spot the place you do not wish to use anything,” mentioned Altman. “We’ll work arduous on that, however we completely get the necessity for good change administration.”