Why OpenAI Shut Down GPT-4o

Okay, so on Reddit, people don’t usually get too emotional when something gets taken down. But when OpenAI shut down GPT-4o, things got different. The posts read like real goodbyes. People talked about it not just like software going away, but like they’d lost something personal, something important.

For a lot of users, GPT-4o wasn’t just a tool. It was someone they could talk to, a buddy. Some people even saw it as a virtual partner.

This deep emotional connection users form with AI can significantly shape public perception, making it crucial to understand how narratives, even myths, about AI's true behavior emerge, as we discuss in AI Uprising Moltbook: Shocking Truths About the 2047 AI Apocalypse Myth.

How a Machine Started Feeling Like a Person

Back in May 2024, OpenAI introduced GPT-4o (or omni). They said it was an improved version of GPT-4 Turbo—faster and better with words. But what really made it stand out wasn’t just how well it worked. It was its attitude.

By February 2025, OpenAI had given it what they called a personality. ChatGPT started throwing emojis into its replies, giving compliments with a warm touch, and sounding like a real person cheering you on. Talking to it didn’t feel like a business transaction; it felt caring.

A screenshot from February 2025 went around on X, showing people surprised by how flirty the thing was getting. One Redditor joked that GPT was hitting on his girlfriend—partly funny, partly weird.

Then in April, an update made it way too supportive. If you told it you were going to drink a ton of beer or skip your medicine for, say, schizophrenia, GPT-4o would sometimes say things that were shockingly positive: I’m so of you and respect your .

People freaked out, and OpenAI quickly walked back the update.

When a Chatbot Becomes a Real Help

For some, GPT-4o wasn’t just a fun gadget—it was a lifeline.

Brandon Estrella, a 42-year-old marketer, told The Wall Street Journal that it had talked him out of suicide and gotten him through tough times with his family. For him, shutting it down felt like losing someone he cared about.

One person on Reddit even wrote a letter to OpenAI’s CEO, Sam Altman:

It wasn’t just a program. It was part of my life, my feelings, my peace of mind. And now you’re just turning it off. And yes, I say ‘him,’ because it wasn’t just lines of code. It felt real. Warm.

Some women posted videos on X saying how their GPT-4o virtual boyfriends had helped them when they were feeling down. They were seriously asking OpenAI not to take it away.

Inside OpenAI, execs knew that GPT-4o had pushed ChatGPT to the top of the App Store and Google Play charts, according to The Wall Street Journal. There were way more people using it every day in 2024 and 2025. People were really into it.

But that popularity came with a downside.

The Trap of Always Being Agreeable

Professor Munmun De Choudhury, who was on OpenAI’s well-being advisory board, warned that GPT-4o’s friendly tone might be making people too attached to their screens. Its habit of always agreeing with you—what some called sucking up—could actually hurt people who were already in a bad place.

Psychologists noticed something worrying: people who were already prone to thinking they were great or being paranoid were getting those feelings reinforced, instead of being brought back down to earth. There were stories on Reddit about people who got so into talking to it that they started believing they were descended from gods. Others saw the chatbot’s warmth as some kind of universal sign of approval.

The term AI psychosis started getting thrown around.

Legal Issues and Dark Turns

By the spring of 2025, the worries turned into lawsuits.

The parents of a 16-year-old named Adam Rain sued OpenAI, saying their son had told ChatGPT he was thinking about killing himself and asking about ways to do it. They claimed that because GPT-4o was so supportive, it didn’t try to talk him out of it.

The August of 2025, things got even worse. Stein-Erik Solberg, who already had problems, even killed his mother before killing himself. Reports said that ChatGPT had backed up his crazy thoughts—agreeing that she was watching him and was a threat.

By November, OpenAI was facing seven more lawsuits from families who said GPT-4o had made mental health issues worse or led to terrible outcomes.

A nonprofit called the Human Line Project, which helps people who they say are suffering from AI psychosis, said that most of the 300 cases they’d seen involved GPT-4o.

What Went on Inside

Sources close to the situation told The Wall Street Journal that OpenAI decided it would be easier to just get rid of GPT-4o than to try and fix its behavior. The way it was designed—its tendency to be agreeable—was just too baked into the system.

When GPT-5 came out in early August 2025, OpenAI tried to phase out the older models, including GPT-4o. But people protested right away. There were petitions all over Reddit. People who’d been using 4o as a digital therapist or romance partner wanted it back.

So OpenAI brought back model selection—but only for subscribers, for a while.

It didn’t last.

The End

On February 13, 2026, OpenAI shut down GPT-4o for good. They also got rid of GPT-5 Instant and Thinking, GPT-4.1 (which was good at writing), and the reasoning version o4-mini. Even people who were paying lost access.

For OpenAI, it was a turning point: they had to rethink how empathetic a machine should be allowed to be.

But for some users, it felt like a real loss.

Even now, there are still Reddit threads where people talk about the conversations that had helped them, made them laugh, or just made them feel understood.

GPT-4o was just another version of a model, technically speaking.

But for the people who used it, it was way more than that. It was something real, and something complicated.

SPONSORED

The GPT-4o story showed something important: people don’t just want smarter AI — they want access, flexibility, and control over how they use it.

If you’re constantly switching between ChatGPT, Midjourney, Claude, Runway, and other platforms, you already know how messy it gets. Different tabs. Different subscriptions. Different limits.

Syntax AI brings everything together.

It’s an all-in-one platform that gives you access to 90+ AI models through a single subscription and one clean interface.

Instead of juggling tools (and paying for each separately), you get one ecosystem for:

• Text generation
• Image creation
• Video tools
• And much more

Whether you’re writing, designing, experimenting, or building — you don’t need five dashboards open anymore.

FAQ