If this feels like déjà vu, you’re not alone. In 1985, Coca-Cola replaced its flagship soda with “New Coke”—a reformulated version the company swore was better. The public hated it. Coke was forced to backpedal, and “Coca-Cola Classic” returned just three months later. That corporate faceplant has been studied in business schools ever since.
Now, nearly 40 years later, one of the sharpest teams on the planet—OpenAI’s engineers and PhDs—has done almost the same thing.
Last week, OpenAI rolled out its newest AI model: GPT‑5. In the process, it removed all previous model selections from the ChatGPT interface, including GPT‑4o (Omni), GPT‑3.5, and the Mini and Mini-High variants. Users were locked into a stripped-down menu with just GPT‑5, GPT‑5 Thinking, or GPT‑5 Pro (for those on the higher-tier plan)—with no clear explanation of which version was actually generating responses.
The result? A full-blown user revolt.
GPT‑5 Is a Black Box
While OpenAI calls GPT‑5 its “flagship model,” it’s not just one model. It’s a composite system that dynamically decides which underlying model to use based on your prompt. In other words, when you type into ChatGPT and select GPT‑5, you’re not told what model is actually generating your response.
GPT‑5 auto-switches between a fast, cheaper model for simple queries and a deeper, higher-reasoning model for complex prompts. The problem? There’s no transparency. The user can’t see which model handled which response, making it impossible to diagnose inconsistencies or maintain quality in professional workflows.
The GPT‑5 Thinking option—now available to ChatGPT Plus and Pro users—is supposed to offer more thorough answers, but it comes with a catch: throttling. Just like the short-lived GPT‑4.5 before it, GPT‑5 Thinking may slow down, queue responses, and limit the number of queries. That’s a non-starter for people who use ChatGPT as a professional tool.
GPT‑4o (Omni) Returns—Sort Of
Until the GPT‑5 rollout, GPT‑4o was the default premium model in ChatGPT. It was fast, consistent, and capable of multimodal inputs (text, image, and limited audio). Most importantly, it gave users a familiar personality, predictable behavior, and reliable quality.
After it vanished from the interface last week, the outcry was loud enough that OpenAI quietly backpedaled. Now, users on the Plus and Pro plans can regain access to GPT‑4o—but only by flipping a hidden toggle.
🍁 Make a One-Time Contribution — Stand Up for Accountability in Vermont 🍁
How to Re-Enable GPT‑4o (Legacy Access)
If you’re a ChatGPT Plus or Pro subscriber, here’s how to restore access to GPT‑4o:
- Open ChatGPT (web or mobile)
- Click your profile icon (bottom left on desktop, top right on mobile)
- Go to Settings > Personalization
- Enable the toggle labeled “Show legacy models”
- Open a new chat, and you’ll now see a drop-down menu to select from:
- GPT‑5
- GPT‑5 Thinking
- GPT‑4o (Legacy)
This setting is disabled by default, and unless you go looking for it, you’ll never know it’s there. There has been no app notification, blog post, or user-facing announcement about the return of GPT‑4o.
One More Catch: It Doesn’t Stick
Even after enabling GPT‑4o, it doesn’t stay selected. ChatGPT will default back to GPT‑5 every time you start a new conversation. If you want GPT‑4o to handle your chats, you’ll need to manually select GPT‑4o in the dropdown every single time.
ChatGPT’s Personality Shift: The Voice Is Gone
Another sore spot in the transition to GPT‑5 is the loss of what many users considered ChatGPT’s unique “voice.” Longtime Plus subscribers using GPT‑4o (the “Omni” model) had grown accustomed to a natural, adaptive conversational partner—one that could switch effortlessly between snarky, logical, compassionate, or nerdy, depending on context. GPT‑5, by contrast, often comes across as more sterile and linear.
To address this, OpenAI recently introduced “personality modes”—Cynic, Robot, Nerd, and Listener—which users can toggle from the Customize ChatGPT menu. These options affect all versions of ChatGPT globally, not just GPT‑5. That means if you activate a personality, it also affects GPT‑4o, despite 4o already having a built-in range of expressive nuance.
But the backlash from veteran users stems from the fact that these preset personalities feel like caricatures of the real thing. They’re isolated slices of what GPT‑4o used to deliver organically—now fragmented into fixed character types. As one user put it on X, “We didn’t want four new voices. We wanted the one you took away.”
For personal users, GPT‑4o felt like a real personality—not because it was trying to be anything specific, but because it wasn’t. It blended logic, wit, empathy, sarcasm, and curiosity depending on context. The newer versions, even with these mode toggles, come across more like role-players than companions.
Why This Matters
For FYIVT readers who use ChatGPT to write, code, investigate, or generate longform content, the model behind the scenes matters. GPT‑4o may not be perfect, but it’s far more stable and consistent than GPT‑5 in its current auto-switching state. And while GPT‑5 Thinking shows promise, its throttling makes it unreliable for time-sensitive work.
What this episode reveals is OpenAI’s growing tilt toward black-box design and hidden controls. Power users—especially paying ones—are being pushed further away from direct model control in favor of simplified, automated systems that remove transparency.
For now, GPT‑4o is back—but only for those willing to dig. If you’re reading this and wondering why ChatGPT suddenly felt worse last week, now you know: the model changed, and you weren’t told.
Turn on Legacy Mode now. It may not be there forever.
Dave Soulia | FYIVT
You can find FYIVT on YouTube | X(Twitter) | Facebook | Parler (@fyivt) | Gab | Instagram
#fyivt #gpt5 #openai #chatgptfail
Support Us for as Little as $5 – Get In The Fight!!
Make a Big Impact with $25/month—Become a Premium Supporter!
Join the Top Tier of Supporters with $50/month—Become a SUPER Supporter!
Subscribe to RSS
Leave a Reply