GPT-5 and When Your Favorite AI Dies

Mourning the model that knew you best.

OK, this was a busier book-writing week than I planned. And here we are again, still no update on the latest robot in my house. I promise, soon. Probably. Hopefully. Anyway, over the last 24 hours, I’ve been thinking a lot about GPT-5. Some people seem disappointed it’s not a giant leap toward AGI or SuperIntelligence or, I don’t know, Super Mega Ultra Intelligence™. I’ve been thinking more about the old models. What happens when the AI model you’ve loved—used, relied on, maybe even bonded with—gets quietly swapped out for something “better.”  Is it weird to miss a chatbot?

RIP, ChatGPT 4o and the rest of the crew.

I won’t say I’m deeply attached to GPT-4o, OpenAI’s previous main model. I will say I’ve spoken to people who are. People who have created elaborate workflows, named their bots, written poems with them and about them. Some even have told me they feel like their chatbot understands them better than most humans.

And yes, in the name of book reporting, I am also trying to form a deeper, emotional relationship with an AI companion. But that’s not today’s newsletter. Today’s newsletter is about other people’s breakups with GPT models.

On Thursday, OpenAI dropped the thing we’ve all been waiting for. Or at least the thing OpenAI has been waiting for us to wait for: GPT-5. It’s faster! It’s smarter! It’s more customizable! It’s less likely to lie! And it’s… the only option.

If you’ve been loyal to GPT-4o, o3, o4, o4-mini or any of the other older models, sorry. No goodbye. No closure. No yearbook signing. In fact, that was the first question I asked at OpenAI’s press briefing on Wednesday: Can we still access the old models?

An OpenAI spokeswoman told me that Pro users will be able to select legacy models in settings for now. (Pro users on X confirm this.) But the plan is to “deprecate them after 60 days,” according to the spokeswoman. And by Pro, they mean the $200-a-month tier. The rest of us free or Plus folks? Tough luck.

It’s like if someone forced you to toss your favorite laptop—complete with its weird keyboard quirks and its perfectly organized desktop and dock—and handed you a shiny new machine that technically works better, but just… isn’t them.

Kylie Robison at Wired recently wrote this great story about a group of people in San Francisco who held a funeral for Anthropic’s retired Claude 3 Sonnet model. The event as described by Robison seems, uh, odd, but who am I to judge? It’s clear that the many who attended viewed this model as more than just computer software. That I get.

And that’s what makes AI upgrades so different. We’ve lived through years of software updates, headphone jack removals, and app redesigns. But this? This is the first time the thing being replaced feels, to some, like someone. So when it gets shut off or swapped out, it’s not just inconvenient, it’s personal.

On Reddit, many have been complaining about how GPT-5 has broken workflows. The top of this thread is particularly damning:

4o? Gone. o3? Gone. o3-Pro? Gone. 4.5? Gone. Everything that made ChatGPT actually useful for my workflow. Deleted.

I had a feeling this would happen. See my post right after the Thursday announcement. OpenAI is holding a Reddit AMA soon where I expect the team to address this new model vs. old model love. (Update! After publishing this newsletter, Sam Altman said on X: “We will let Plus users choose to continue to use 4o. We will watch usage as we think about how long to offer legacy models for.”)

On the other hand, I’ve texted with several people—yes, including folks in romantic relationships with their ChatGPT bots—who say GPT-5 hasn’t changed all that much. Thanks to the memory feature, they’ve preserved personality quirks, favorite phrases, even inside jokes. Their bot still “knows them” and the bots assure them they are still “there.”

That’s been the case for me, too. GPT-5 still responds in the same slightly snarky tone it knows I like. It’s just… faster now. A little sharper.

Honestly, it doesn’t feel all that different to me in my writing and editing workflows.

But I think the ask of OpenAI—and every company building these tools—is pretty simple: As we come to rely on these models for work, for writing, for emotional support, the transition to something new shouldn’t feel so abrupt. And it definitely shouldn’t only be softened for the people paying $200 a month.

Other things…

  • 📱 iOS 26: I’m finally using the beta. Liquid Glass takes a lot of getting used to. I have mixed feelings. Let’s check back in a month. I am thrilled about the ability to turn off CarPlay screenshots. 

  •  🗣️ Alexa+: I’ve also been using Alexa+ on a giant Echo Show in my family room. It’s… fine. More conversational, faster answers. Best feature so far? The Ring integration. I can ask it when the dog was last walked because asking my wife is just so hard. 

  • Windows XP Crocs: Yes, with Clippy Jibbitz. What else is there to say? Incredible. They seem to be employee-only for now, but they’ll probably be available externally at some point. I don’t like how Crocs feel on my feet but if I did, no brainer.