“It will be a great help to experts, but it will definitely not replace them.” Ukrainian IT professionals share their impressions of working with GPT-5

Mykhailo Patsan, investor, entrepreneur, expert in the fields of cryptocurrencies EdTech, FinTech, Web3, AI

GPT-5 is really cool. Tested. Impressed. We are updating our customers' AI Agents. 3D Tanks made with one prompt.

As Sam Altman said, «You have a PhD-level assistant in your pocket,» and it really is. More empathetic, more intelligent, and more precise.

I was skeptical, thinking «another model that performs a little better than GPT-4,» but no — this is a new era. Why?

GPT-5 is built using agentic approaches and actually combines several sub-agents. It can operate in three modes: basic for quick answers, thinking for more complex ones, and «concrete thinking» for deep analysis.

It already works in ChatGPT. And yes, now there will be only one GPT-5 model, nothing more.

There are already four models available for developers: GPT-5, GPT-5-mini, GPT-5-nano, and GPT-5-Chat. They are cheaper than GPT-4.1.

Yevhen Oleksenko, QA Manual Engineer, Web and Web3 specialist

Well, I decided to use GPT-5 a little and was genuinely impressed by the speed and quality of work)

The first thing that came to mind was to create a beautiful, useful, and truly interactive Git reference page for QA engineers. I like to create quick tools/references for myself that I can access quickly. This optimizes time and has a positive effect on my work.

A few clarification prompts and we have a modern page with a space 3D background on Three.js, neon accents, parallax, instant search, one-click command copying, and a theme switcher — all in one HTML file, without complex assemblies.

The most enjoyable part was how smoothly the process went: I described the idea, added a few minor edits («I want a futuristic look,» «add all the modern Git commands,» «let’s have 3D stardust»), and after a few iterations I got a ready-made tool that is really convenient to use every day.

AI here acted not just as a «code generator,» but as a co-developer who quickly turns an idea into a working product .

Volodymyr Bandura, co-founder of Allbionics

In short, I think the model is cool and a good business proposition. I will write in detail why and for what purposes. With one important caveat…

…it would be a big mistake for someone to run through 2-3 prompts and make some generalizations. These models are very complex artifacts, some kind of archives of human knowledge. We will be able to give a real assessment soon and after detailed testing. In our real tasks, in comparisons with other models…

…GPT5 can be a full-fledged and permanent «team member» and has something to add. And it detects very subtle problems in the code. Moreover, the more difficult the task, the more interesting things it can add.

And what is important for practice — it also works quickly. Because there was such a model gpt o4 mini. It seems to be good according to tests, but in high mode it is such a brake — that with it the work goes 10 times slower than with Claude, and who needs that…

…Well, I’ll note — if you were expecting AGI, superintelligence and terminators to suddenly crawl out of the screen, then this is not about that. The model makes sense. Very capable. It will be a great help to experts, but it definitely won’t replace them, I’ve written a lot about this. A good business proposition. It’s cool that even for free users will be able to ask several questions a day. Not radically better than key competitors.

And if this is the Singularity, then it is still very gentle))

Serhiy Didkovsky, PR strategist at Adwise.agency

GPT-5 really thinks deeply. Of course, it (like other models) needs to be tuned, trained, and understood to work, etc.

Artificial intelligence is as important to evolution and technological progress as the Internet itself.

Vasyl Zadvorny, AI enthusiast

Yes, I already ran GPT 5.

No, it cannot draw conclusions yet, it requires quite a long period of testing on cases.

Yes, the application itself is very powerful.

No, it doesn’t think.

Kristina Nikolaeva, digital content author

Yesterday’s presentation of GPT-5 wasn’t really surprising, to be honest, because many of the new features could have been predicted if you followed OpenAI’s updates over the past six months and thought about where they were leading. I’ll write about the new capabilities of this model separately.

But there was one block that I haven’t been able to let go of since last night — GPT-5 for Health. It was in it that I felt the most how AI updates can improve the quality of life of people in the world, contrary to all the apocalyptic sentiments.

GPT-5 can:

• analyze symptoms, medical records, images, and language simultaneously;
• speak in a calm, delicate voice, clarifying important points;
• to be there between visits when questions arise but there is no one to discuss them with.

It does not replace a doctor. But it helps you remember important things, explains them again (or 5 more times, if necessary), in simple language, and reduces anxiety while waiting for a diagnosis or procedure.

And for people with serious health problems and their loved ones, the opportunity to discuss their condition, test results, diagnosis details, etc. at any time can be a lifeline to avoid losing common sense and hope.

GPT-5 is already being tested in clinics as an assistant. In a world where millions of people cannot receive qualified medical care in a timely manner, such solutions could become critically important.

Because sometimes it’s enough for someone to just be there and explain.

The author of the Telegram channel " Ooo Neural Network »

Okay, I have to admit it: the GPT-5 update is a failure and a scam.

No, I didn’t expect anything more than what was shown, but the problem is that we didn’t get what was shown.

By points:

1. GPT-5 has broken routing and almost all of the time my requests go to GPT-5-mini;
2. They severely nerfed the model’s reflection properties: even extremely complex tasks, GPT-5 ponders for 15 seconds and then shoots me 1000+ lines of code that doesn’t work. To understand: for Plus users, Thinking is ~3 times weaker than it could be;
3. GPT-5 is worse than previous models in every way. It is not as pleasant to talk to as GPT-4o, not as creative as GPT-4.5, and not as smart as OpenAI o3 (see point 2).

There’s an AMA planned on Reddit with OpenAI’s top people. Every first post is about how GPT-5 sucks and asking for the old models to be brought back.

I don’t have much faith in Google yet, but I’m starting to look forward to Gemini 3 a little more.

Ruslan Khomenko, digital content author

My personal opinion is that progress in AI is very rapid, but still gradual, and the decisive factor in business should probably not be the creation of the most powerful model, but the ability to widely implement the technology economically. GPT-5 fits this criterion quite well. In the scenario of scaling models, China has several strategic advantages, including a developed energy infrastructure and a proven ability to quickly scale new technologies.