Photo by Erwan Hesry on Unsplash
OpenAI’s latest model, GPT-5, landed with high expectations—and a few missteps that got people talking. From users calling it “dumber” than before to a now-infamous botched chart during the launch, things didn’t go exactly as planned. But CEO Sam Altman isn’t hiding from the hiccups. In fact, he’s answering questions, owning up to mistakes, and promising improvements.
“GPT-5 Felt Dumber”? Here’s Why
On launch day, many users noticed that GPT-5 just wasn’t performing the way they’d hoped. It felt slower, less accurate, and – as many put it – just not as sharp as the previous model, GPT-4o.
Altman explained that wasn’t entirely GPT-5’s fault. The real issue? A faulty “real-time router,” a new system designed to decide which AI model should respond to a query. It’s meant to balance speed and depth by switching models depending on the prompt. But the router was struggling out of the gate.
“GPT-5 will seem smarter starting today,” Altman said, admitting that a service event (called a “sev”) knocked the auto-switcher offline for part of the rollout day. That led to the model appearing less intelligent than it actually is.
Better routing logic is already in the works. And importantly, we’ll soon have more visibility into which model is answering—so you’ll know exactly what’s behind your AI conversations.
Photo by Markus Spiske on Unsplash
Users Clamor for GPT-4o… and OpenAI Might Listen
A big chunk of user feedback was about missing GPT-4o. Many felt the older model was quicker or more useful in specific tasks. And they weren’t quiet about it.
Altman didn’t dismiss the idea. Instead, he said, “We are looking into letting Plus users continue to use 4o. We are trying to gather more data on the tradeoffs.” So, while it’s not confirmed, don’t be surprised if GPT-4o makes a comeback for paying users.
He also promised that the company is “going to double rate limits for Plus users” during the rollout. That means more queries, more experimenting, and more ways to get a feel for GPT-5 before deciding how it fits into your workflow.
Photo by Pawel Czerwinski on Unsplash
That Awkward Chart Moment
Of course, it wouldn’t be a product launch without at least one meme-worthy moment. In this case, it was the highly inaccurate chart from OpenAI’s GPT-5 presentation. The bar graph showed a much lower benchmark score with a ridiculously taller bar—something that caught the internet’s attention fast.
Altman later called it a “mega chart screwup” and acknowledged the mistake. The corrected charts in blog posts were accurate, but the damage was done. One reviewer even joked that turning data into a table was a good example of GPT-5’s early flaws.
It was a small but telling reminder: shiny features don’t mean much if the basics aren’t solid.
What Now?
Despite the rocky start, OpenAI is responding quickly—and publicly. That’s more than can be said for some tech launches.
Here’s what’s coming soon:
- A more reliable real-time router
- Increased transparency on which model is responding to queries
- A possible return of GPT-4o for Plus users
- Doubled rate limits for Plus subscribers
Altman ended his Q&A by reaffirming their roadmap: “We will continue to work to get things stable and will keep listening to feedback.”
So if GPT-5 has felt off to you, you’re not alone—but things are changing, and fast.
Stay tuned. GPT-5 is still finding its feet. And OpenAI seems ready to walk the talk.
Keywords: GPT-5, OpenAI, AI models, Sam Altman, GPT-4o
