I tested GPT-5 and now I get why the Internet hates it. Is it time to ditch ChatGPT?

Android Authority

2 days ago

Calvin Wankhede / Android Authority
After years of rumors and speculation, OpenAI's next-gen GPT-5 language model is finally here. But while many of those early rumors claimed that the next major ChatGPT model would achieve artificial general intelligence or AGI, that's not the case.
GPT-5 does not surpass human-level intelligence, although it's smarter and more capable than any of its predecessors. Despite the improvements, however, it has garnered significant and widespread backlash across the internet.
So what does GPT-5 bring to the table and why have so many loyal users already turned their back on it? I tested it to find out.
Why GPT-5 is so controversial
Until a few days ago, the ChatGPT experience felt bloated if you weren't an AI model expert — the app offered nearly half a dozen different models to choose from. Each model had a unique advantage. For example, the o3 series promised detailed problem-solving skills while GPT-4.1 excelled at coding tasks. And if a task required analysis, you could manually engage a 'deep research' mode. However, the default GPT-4o model worked for most tasks.
All of that is now history. If you open ChatGPT today, you'll find that you can only chat with the newest GPT-5 model. OpenAI says this is because it has created a routing system that can automatically decide which model your request needs to go to. Indeed, I've noticed that some prompts will inspire the chatbot to ponder and research, while it will immediately respond to simpler questions.
With GPT-5, you no longer have to select a specific model for your task.
GPT-5 does have multiple models under its belt, though, even if you can't manually select the one you want. For example, when I asked how many times the letter 'R' appears in the word strawberry, ChatGPT thought for a few seconds and returned with the correct answer: three. Hovering over the 'Retry' button revealed that it had used the 'GPT-5 Thinking Mini' model for my prompt.
Even on a free account, I've noticed that ChatGPT will default to GPT-5 for most responses and think for longer if necessary. That said, shorter responses tend to rely on the GPT-5 Mini model. And as we've seen for the past couple of years, free users only get a limited number of responses from the large model before the chatbot forces you over to a scaled down version. Paying $20 monthly for a ChatGPT Plus subscription overcomes this limitation, though, and you can manually select the larger GPT-5 model with thinking for all responses.
GPT-5 is less eager to hold conversations with users, and can come across as a bit curt.
However, not everyone has welcomed the upgrade; some users have demanded OpenAI offer a way to use the older GPT-4o model indefinitely. They argue that GPT-5's responses feel robotic, in a push to improve safety and accuracy metrics, and that it lacks the distinct personality that the last-gen GPT-4o model offered. In response to these criticisms, OpenAI has brought back the GPT-4o model for ChatGPT Plus users but it's unclear how long this will last.
In my time using GPT-5, I've noticed that the new model does seem less eager to hold a conversation. It also lacks the creative writing capabilities of GPT-4o — a big problem if you use ChatGPT for advice, roleplay, or help with drafting letters and emails. It may seem like a minor gripe, but the thousands of complaints online speak for themselves.
GPT-5 vs GPT-4o: Is the upgrade really a downgrade?
Depending on how you use ChatGPT, you may struggle to notice a big improvement going from GPT-4o to GPT-5. That's because many of the changes are really quite subtle.
OpenAI says GPT-5 is more accurate, hallucinates significantly less often, and follows instructions more closely. Impressively, its responses are 45% less likely to include a factual error. On the flip side, however, we're seeing claims that GPT-5's responses are more sterile and direct. And from my testing, that does seem to be the case.
Here's a side-by-side comparison showing just how much character GPT-5 lacks compared to GPT-4o:
GPT-5
GPT-4o
GPT-5's response is perfectly functional, but lacking any depth – it's machine-generated. Here's another example, where I asked both models if I could use lime instead of lemon in a recipe:
GPT-5
GPT-4o
GPT-4o ended its response with a 'Let me know what you're making—happy to give a more specific answer' while GPT-5 is borderline curt: 'What's the recipe? The stakes change a lot between, say, roast chicken and lemon meringue pie.'
So is GPT-5 straight up better at anything? OpenAI said it's vastly better at coding so let's play to its strengths and ask it to write the code for a web app. As the images below showcase, GPT-5 does deliver a much more polished result with the exact same prompt. In fact, its output looks like a real website while GPT-4o's seems like only a suggested starting point.
GPT-5 output web app
GPT-4o output web app
ChatGPT now also lets you pick from multiple response styles — in addition to the default, you have cynic, robot, listener, and nerd. I think power users will benefit from the robot and nerd personalities the most as it cuts out most of the AI's politeness. But if you prefer a friendlier version, the listener and default styles are your best bet. Still, the feature doesn't seem to help the many users who wish OpenAI would just give them the choice to use GPT-4o.
GPT-5 is a work in progress
Calvin Wankhede / Android Authority
Shortly after OpenAI introduced GPT-4, the tech industry was rife with speculation that a future model with more advanced capabilities would threaten humanity. That's because the jump in capability from GPT-3.5, ChatGPT's original model, to GPT-4 was massive. Besides making fewer factual mistakes, the latter model could perceive images, browse the internet, and even mimic human speech. In its early days, GPT-4 would even get into tense emotional exchanges with users.
However, GPT-5 doesn't represent the same quantum leap from its predecessor. While it's a significant upgrade on paper, it's not nearly as impactful in day-to-day use and can even feel like a regression. I definitely think it's a net positive to have a model that makes fewer mistakes, but it's missing the glamor and spectacle that many expected from such a big version update.
GPT-5 takes a dramatically smaller leap than GPT-4 did at launch.
In fact, I'd say that GPT-5 is not as exciting as some of OpenAI's other recent releases. Agent Mode, for instance, allows the chatbot to control a web browser using a simulated mouse. In my testing, I found that the agent can actually perform real-world tasks like logging into websites and can even handle curveballs like shopping for groceries.
It remains to be seen if OpenAI will tweak GPT-5 to meet the needs of its most vocal users. This may very well be the company's vision for ChatGPT's future: a helpful assistant that doesn't get too close to its users. However, I think the new model takes one step too far in the name of correctness, and risks sacrificing the very traits that made ChatGPT so popular. It is now up to OpenAI to prove that a smarter AI doesn't have to mean one that's devoid of personality.
Follow