When GPT-5 was released on August 7th, I was concerned. I’m an avid user of AI for my clients—though only on client request—and I use it for occasional personal projects as well.

I won’t be on this website, unless explicitly mentioned, as I’m trying to showcase my abilities and skills. That, and having fun publishing short stories and content.

But there is one caveat here, because I’m not a professional coder. Though I know it’s probably been a detriment to my abilities, I’ve refrained from learning to code. I used to know how to code, if you consider HTML & CSS to be coding, but those skills have been lost to time and memory.

I no longer recall my abilities in these areas, and don’t have the time to relearn these skills. And so, when GPT-5 was powered up, I was excited to test it for one purpose—upgrading my website.

The release

And so, I powered up my OpenAI dashboard, which I use via the OpenAI Playground, and got to testing.

The results were horrifying. As a fun test, I decided to feed the AI a prompt to draft a poem. Though it’s a reasoning model, and not explicitly built for naturally flowing user content, I assumed at the very least it would be able to make something rhyme.

So, consider me disappointed when it failed at this. Not mildly, either—the AI completely fudged every rhyme it attempted to write.

The disappointment

After all that time waiting for the release of the latest OpenAI model, I was shocked and put off, to say the least. Why release such a promising model, only to ignore such obvious shortfalls?

That’s not to say that I was one of those individuals who’d been spreading promises of AGI or some other fanciful system. I wasn’t expecting much, but even still, I felt a sense of disappointment.

The upgrade

Of course, I headed to bed that night wondering if I ought to go with a more powerful model from a competing organization.

I wouldn’t have minded the release of GPT-5 by itself, but the looming threat of the other models being discontinued was very real. I use OpenAI to streamline various facets of my work, and losing access to credible models would have been heartbreaking.

Well, luckily, it didn’t last. I’m not sure why—maybe I just prompted it incorrectly, or maybe there was some mysterious upgrade that took place between the two days. But in the morning, when I tested it again, the program was actually able to rhyme.

Results

Simply testing it at the moment I’m writing this article, on the 12th, the system seems to have improved from my first experience. Even using the deep-thinking, fully-fledged GPT-5 model rather than the minimal chat version, it actually manages to produce something tenable.

I prompted it, “Write a short poem that rhymes.”

And here is the result:

Between the trees, a silver thread of light Winds through the whispers, stitching night to day; The world holds still to catch its breath in flight, Then laughs in leaves and softly slips away.

Beautiful, right? At the very least, there’s a competent rhyme scheme here, which is more than I could have said originally.

But of course, that’s not where the real power of this AI model lies.

The real power

So what’s the real use of GPT-5? You likely already know, but I’ll list out some of the biggest uses I’ve found:

  • Coding. GPT-5 excels at coding, capable of fixing errors and very rarely makes errors itself.
  • Improved recall within context windows. Other previous models struggle with recalling and tend to hallucinate the more you converse with them. Not GPT-5 nearly as much.
  • More customization. When it comes to customizing your responses, you now have access—via the Playground dashboard—Verbosity, as well as an extra level for Effort titled “minimal”.
  • Verbosity. Speaking of Verbosity, that’s a huge benefit to this new model. It no longer struggles to write long content. It’ll churn out entire pages of code without even questioning my instructions or trying to take shortcuts.

So though I was initially disappointed, I feel as if the model itself has improved, and I’ve begun to recognize its strengths and weaknesses.

Do I think that OpenAI should have perhaps vetted it better? If there’s any credence to my claims, absolutely.

Conclusion

But as far as it goes, GPT-5 is a surprisingly solid model. It’s not perfect, but it goes a long way in terms of reaching the level of intelligence required for a “true” AI system that adheres to instructions at a level previously unseen.

To infinity, but hopefully not beyond!