I recently tried ChatGPT-5 for over a week, and I have to say, I was impressed with its accuracy and ability to keep up with my pace. It even threw in some sarcastic remarks, which I appreciated. However, when it came to coding, I realized that it just wasn’t up to par with my old setup.
I used to have a dream team of models: o4-mini high for heavy coding jobs, o3 for breaking things down into plain language, and 4o for quick lower-level questions. With ChatGPT-5, I tried regular mode and let it choose when to switch into thinking mode. But it would often switch back to regular mode in the middle of a task, losing track of what we were doing. I had to force it back into thinking mode and even then, it would forget what we were working on.
To make matters worse, it struggled to hold context. I had a separate chat about an overwatered aloe plant I was trying to save, and when I went back to that chat to give an update, ChatGPT-5 confirmed the steps and then said, ‘After that, your snake plant should recover nicely.’ I was shocked that it couldn’t even remember the type of plant we were discussing, especially since I had mentioned it earlier in the conversation.
I was so frustrated that I switched back to 4o and asked it what to do next. To my surprise, it remembered the name of my aloe plant, Bean, and gave me the correct steps to take. It was clear that 4o was still the superior model when it came to holding context and remembering important details.
In the end, I realized that ChatGPT-5 just wasn’t ready to replace my trusty old models. Maybe one day it’ll get there, but for now, I’m sticking with what works.