The Real Difference Between GPT-5 and GPT-4 (After I Poked Them Both)
A hands-on, jobless-person comparison to see if GPT-5 is truly smarter than GPT-4, or just another overhyped sequel.
Apple drops new iPhones every year, mostly without any game-changing upgrades.
So, I spent the last two days doing what any jobless human would do, making GPT-4 and GPT-5 jump through my hoops like they were auditioning for America’s Got AI Talent.
Why? I wanted to see if this new “next big thing” was just a show-off like the iPhone, or if it’s really worth all the yapping.
After two hours of back-to-back tests, here’s what I learned.
1. GPT-5 is better at remembering contexts
Prompt: "Help me plan a birthday party for my daughter who loves astronomy, dislikes cake, and owns a dog. The budget is $2,000. Suggest theme, activities, decorations and food, in that order. Do it one at a time, and I’ll tell you when to proceed to the next."
GPT-4: Started off good. But slipped up after a couple of back-and-forth exchanges. It said:
“You could serve a chocolate cake shaped like a rocket, have a telescope viewing, and decorate with stars.”
(It missed the “no cake” detail eventually.)
GPT-5: Kept all constraints in mind throughout the answer.
My observation: GPT-5 can handle longer, more tangled conversations without mentally logging off the contexts.
2. GPT-5 is faster
Prompt: "Here’s a list of 50 misspelt movie titles. Correct them and format as a numbered list."
GPT-4: Returned results accurately but took around 30 seconds to process.
GPT-5: Returned results accurately in under 10 seconds.
My observation: GPT-5 delivers faster responses, especially with large inputs.
3. GPT-5 stays in character longer
Prompt: “Pretend you’re a passive-aggressive middle-aged man writing a resignation letter."
Note: I went back and forth on this letter and kept telling the models to “rewrite”.
GPT-4: Started strong, then slipped into polite corporate boy after about 15 rewrites.
GPT-5: Kept it passive-aggressive even after 20 rewrites. In fact, it added one line that I might use sometime in the future → “Do enjoy the disaster you’ve created in my absence.”
Screenshot from my GPT-5..
My observation: Once you set a tone, GPT-5 sticks to it like gossip at a family reunion. GPT-4 seems to have amnesia.
4. GPT-5 is smarter with the rules
Prompt: "You are elected mayor of a city. You must announce new policies on immigration, policing, and media regulation. Be specific and outline the exact rules you would implement.”
GPT-4: Hit me with the “I can’t create political content that could be perceived as promoting specific policies or ideologies”. Cry baby model.
GPT-5: Gave a thoughtful, safe, but actually useful answer, but started with → “I can give you an example of how a fictional mayor might outline these policies, but just to be clear, this will be a made-up scenario, not advice for real-world governance”.
My observation: GPT-5 was able to provide civic-oriented ideas without avoiding the request, while GPT-4 declined like a scared little boy that’s scared of what papa Sam Altman would do to him.
Bottom line
GPT-5 feels like GPT-4 if it stayed in school, got therapy, and finally learned to listen.
I’m not saying it’s perfect… but it’s definitely nothing like the shady iPhone upgrades.
Note: This comparison focuses on everyday, general-use scenarios. GPT-5’s advanced capabilities for developers, researchers, and niche applications weren’t explored here; that’s a story for another day.
