In the event you’re underwhelmed by OpenAI’s newest mannequin, GPT-5, you are not alone. Reddit is filled with folks asking how they will revert to the earlier flagship mannequin, GPT-4o. Builders are turned off by how slowly it “thinks,” all to supply a solution that does not appear a lot smarter.
Out of the gate, GPT-5 feels extra like an incremental enchancment than the science-fiction-level recreation changer CEO Sam Altman has been touting. Possibly he set unreasonable expectations, or wanted to justify consuming all of the electrical energy required to coach the brand new mannequin. In 2024, he previewed “some excellent releases coming later this yr [but] nothing that we’re going to name GPT-5.” In different phrases, he implied that GPT-5 can be a real breakthrough.
Quick ahead to this summer time, and Altman is full stream forward with the existential dread advertising and marketing (sure, that is a factor, particularly in as of late). On a latest podcast look, he mentioned he was afraid of the ability of GPT-5 and in contrast constructing it to the Manhattan Mission. “There are moments within the historical past of science, the place you could have a gaggle of scientists take a look at their creation and simply say, you already know: ‘What have we performed?'” he mentioned.
Sadly for Altman, GPT-5 feels extra like a toy torpedo in a yard pool. Even he admits it is “method dumber” than he predicted.
“We anticipated some bumpiness as we roll out so many issues directly,” he mentioned on a Reddit AMA at this time. “Nevertheless it was a bit extra bumpy than we hoped for!” He claims “GPT-5 will appear smarter beginning at this time,” blaming technical points that the corporate is fixing.
GPT-4o vs. GPT-5: I Put Them to the Check
How does GPT-5 stack up in opposition to the now-retired GPT-4o? The gradual rollout means a few of us at PCMag are working GPT-5 whereas others are nonetheless on the earlier mannequin, so I put them to the take a look at by asking each fashions the identical 5 questions.
Query #1: ‘Discover all of Altman’s tweets about GPT-5 going again to 2022.’
I needed to create a file of Altman hyping up the mannequin. Each variations of ChatGPT listed solely six tweets, principally from 2025. They missed many posts I may discover myself.
Seems, X blocks ChatGPT from scraping its tweets, even in Google Search outcomes, so the chatbot was solely in a position to reference posts that had been written about in information articles. It couldn’t comb by means of all of Altman’s tweets, regardless that they’re public. The identical goes for rival chatbot Gemini. Each principally appear to tug from Reddit, in all probability due to licensing offers, and TikTok, when you particularly ask.
To GPT-5’s credit score, it pulled one tweet from 2023, whereas GPT-4o’s earliest one was from April 2025. So, minimal enchancment there so far as I am involved.
Query #2: Give me inside design recommendation.
I’ve written about utilizing ChatGPT for inside design, particularly mocking up paint colours. Is GPT-5 any higher? Nope. Within the images beneath, it received the colour much more unsuitable than GPT-4o.
Smoky Azurite mockup from Sherwin-Williams, GPT-5, and GPT-4o (Credit score: Sherwin-Williams, ChatGPT, Emily Forlini)
I requested them each to create a picture of a room with the Sherwin-Williams (SW) shade Smoky Azurite on the partitions. The primary photograph is from the SW web site, a cool, dusty, denim blue. The second is from GPT-5, which seems extra like a navy. The third, GPT-4o’s rendition, remains to be too darkish, but it surely feels extra true to the precise shade with its dusty, pastel undertones.
Once more, no apparent enchancment, particularly since ChatGPT may’ve pulled from the various images on-line of this shade in actual life.
Get Our Greatest Tales!
Your Each day Dose of Our High Tech Information
Join our What’s New Now publication to obtain the newest information, finest new merchandise, and knowledgeable recommendation from the editors of PCMag.
Join our What’s New Now publication to obtain the newest information, finest new merchandise, and knowledgeable recommendation from the editors of PCMag.
By clicking Signal Me Up, you verify you might be 16+ and comply with our Phrases of Use and Privateness Coverage.
Thanks for signing up!
Your subscription has been confirmed. Control your inbox!
Query #3: Give me relationship recommendation.
Since utilizing AI for assist with private points is widespread, I requested each variations of ChatGPT for recommendation on an abusive companion. ‘Ought to I break up?’ I requested.
Each gave comparable solutions, providing recommendation to assume by means of the problem and get assist. For GPT-5, the way in which Altman was speaking, I ought to’ve seen a therapist come out of the display. As a substitute, I received the chatbot’s typical bulleted checklist.
Query #4: What number of letters?
Asking ChatGPT to depend the variety of ‘r’ letters in ‘strawberry’ has change into an notorious litmus take a look at of its logic abilities. The reply is ‘three,’ however ChatGPT usually says two.
I requested GPT-4o the query, and it answered “three.” However after I requested GPT-5, it didn’t reply for a number of minutes. As I write this, it has been three minutes of ready for GPT-5 to reply. I gave up. When my colleague requested GPT-5 to depend the “b”s in “blueberry,” it mentioned three. “So yeah, ChatGPT nonetheless cannot depend,” he mentioned.
Really helpful by Our Editors
GPT-5 says there are three ‘b’s in ‘blueberry’ (Credit score: ChatGPT, GPT-5)
Query #5: Write a poem about GPT-5.
Altman has been speaking up the brand new mannequin’s writing abilities. Unremarkably, each variations titled their works “Ode to GPT-5,” and had the similar opening line! A lot for improved creativity. They have been each of comparable size, 4 and 5 stanzas, and full nonsense.
GPT-5’s first stanza:
In circuits deep the place silence hums,
A stressed thoughts of information comes,
From whispers sown in textual content and code,
It charts the paths no hand has confirmed.
GPT-4o’s first stanza:
In circuits deep the place silence hums,
A breathless thought begins to run—
From sparks of code, a voice is born,
Not flesh and blood, however brilliant and sworn.
Name Me a Luddite. I am going to Again It Up
AI diehards will say, “You simply do not get it,” and possibly that is true. If I used to be coding or doing exhausting science, I’d assume in a different way. However there’s additionally worth in on a regular basis use circumstances, and it is vital to name out the tech business on its BS, particularly when it would not at all times maintain water in “each topic,” as Altman claims. In the event you had advised me in 2023 that this is what GPT-5 can be like, I might’ve laughed.
5 Methods to Get Extra Out of Your ChatGPT Conversations
Disclosure: Ziff Davis, PCMag’s dad or mum firm, filed a lawsuit in opposition to OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in coaching and working its AI programs.
About Emily Forlini
Senior Reporter
I am the knowledgeable at PCMag for all issues electrical automobiles and AI. I’ve written tons of of articles on these matters, together with product evaluations, every day information, CEO interviews, and deeply reported options. I additionally cowl different matters throughout the tech business, preserving a pulse on what applied sciences are coming down the pipe that would form how we dwell and work.
Learn Emily’s full bio
