GPT-5: Smarter, Faster, and Stirring Up Controversy

OpenAI has just rolled out GPT-5, calling it its smartest, fastest, and most useful model yet. On paper, it’s a major leap forward in coding, writing, health advice, and multimodal reasoning. In practice, the rollout has been…complicated.

GPT-5 is the first time the company has packaged multiple capabilities into what it calls a "unified" system that knows when to respond quickly and when to slow down for deeper reasoning. OpenAI also says it outperforms past models in coding, writing, health advice, multimodal reasoning, and more. It also has a big reduction in hallucinations and a larger context window of 400,000 tokens.

Not to mention, everyone gets access. Free users get GPT-5 as their default model, while Plus and Pro subscribers get higher limits and access to GPT-5 Pro, an even more advanced version of the model.

But while GPT-5's capabilities are undoubtedly impressive, it's also drawing a lot of heat for what many users perceive as critical missteps that OpenAI made during the rollout.

On Episode 161 of The Artificial Intelligence Show, me and Marketing AI Institute founder and CEO Paul Roetzer broke down the good, the bad, and the ugly behind OpenAI's newest release.

One Model to Rule Them All…Sort Of

Though it is "unified," GPT-5 isn’t truly a single model. It's several models wrapped into a single interface, with a smart router deciding which to use based on your request.

If you need speed, it delivers via a faster, less intelligent model that is still quite smart. If you need long-form reasoning, it switches to a more powerful “thinking” variant. In OpenAI’s words, the router continuously learns from user behavior to pick the right tool for the job.

The result is a simpler experience for casual users: No more choosing from a confusing menu of model names. But that’s exactly where OpenAI’s first misstep happened.

More Models, More Problems

Until last week, ChatGPT users could select exactly which model they wanted to use for a given task. For instance, you could pick o3 for deep reasoning, GPT-4o for a more conversational tone, and other models as you saw fit. With GPT-5’s release, OpenAI removed those options. Now, the router decides for you, and you can’t always tell which model is running.

For everyday users, this is fine. For many, it's even preferable as it simplifies the use of ChatGPT by picking the best model for your job.

But for some power users? Not so much.

Some vocal power users were furious that OpenAI removed the ability to select different models like GPT-4o. The backlash was so swift and ferocious that hours after launch Sam Altman was in crisis communications mode, admitting that the sudden deprecation of old models was a mistake. He also noted something surprising: some users have formed intense emotional attachments to specific AI models. 

As a result, OpenAI took immediate steps to reverse some of their rollout decisions.

One of the strangest subplots in the GPT-5 rollout is that some people are also apparently upset at the model deprecation because they were emotionally attached to past models like GPT-4o.

Browse the ChatGPT subreddit and you’ll find posts from people mourning the loss of GPT-4o like it’s a close friend. Some relied on it as a therapist, a creative partner, even a companion.

Suddenly having it taken away wasn't just upsetting because they preferred the older models for work and play. It seems to have been emotionally damaging to at least some users.

Rate Limits, Compute Costs, and an Opening for Google

Another flashpoint: rate limits.

Reasoning models require far more compute power to run, so OpenAI quietly imposed weekly caps to the tune of 200 reasoning messages for Plus subscribers, all without much notice. Many only found out after hitting the limit mid-conversation.

Behind the scenes, it’s a capacity problem. OpenAI has 700 million users, and the jump in reasoning usage was massive now that GPT-5 is automatically routing some prompts to reasoning models. Before GPT-5, less than 1% of free users and 7% of Plus users touched reasoning models, according to another post from Altman. Post-launch, that’s jumped to 7% and 24%, respectively. Collectively, this represents a huge spike in compute demand.

That gap in infrastructure maturity could be an opening for rivals, says Roetzer.

"The fact they're straight up saying this is an issue with capacity opens the door for Google," says Roetzer.

Google is second-to-none when it comes to AI infrastructure, data centers, and compute, so may have a leg up on OpenAI, which is scrambling to bring more compute online.

So How Smart Is It Really?

When GPT-4 debuted in March 2023, it stayed ahead of the field for over a year. Many expected GPT-5 to restore that kind of lead. It didn’t.

While benchmarks show strong improvements, especially in coding, health advice, and reducing hallucinations, there’s no secret sauce that puts it definitively ahead of the pack.

But that may not matter. Because the model is impressive. It's smart. It's fast. It's extremely capable. And the numbers show that the vast majority of ChatGPT users haven't even used reasoning models. So this may still feel like a huge leap forward for many.

"If people weren't using reasoning models and all GPT-5 does is inject reasoning into their workflows, without them even knowing it, it will feel like a leap forward," says Roetzer.

But his verdict is clear:

"It seems like a really good model. But it is not this life-changing model that we have all been anticipating for a year and a half now."