• Industry News, Web Development • 10:47 pm

GPT-5: The Day OpenAI Broke Everything

OpenAI’s forced GPT-5 rollout removed all previous models and left users with a broken, loop-prone AI that can’t handle basic tasks. A cautionary tale of how not to launch a product.

3 min read
GPT-5 blog post featured image

The Day OpenAI Lost Me

Look, I get it. Progress marches on. New models drop, old ones fade. But what OpenAI pulled with GPT-5? That wasn’t progress—it was a hostage situation.

August 7th, 2025. The day they flipped the switch and force-fed us GPT-5, yanking away every previous model like a dealer cutting off supply. No GPT-4. No GPT-4o. Just their shiny new toy that, spoiler alert, is about as sharp as a bowling ball.

My Workflow, Interrupted

Here’s the thing: I’m all-in on Claude Code for actual development work these days. When I need to ship code, architect systems, or debug something gnarly, Claude’s my copilot. But ChatGPT? That was my thinking partner. My brainstorming buddy. The place I’d go to workshop ideas, research topics, or just explore concepts without the pressure of production code.

Professional projects, personal curiosities, random 2 AM thoughts about quantum computing—ChatGPT was where those conversations lived. Was being the operative word.

The Nightmare Begins

The new GPT-5 experience? Picture this: You ask a straightforward question. The model starts answering, then catches itself in some recursive hell-loop, repeating the same three sentences like a broken record. You try to redirect it. It apologizes, starts fresh, then—boom—right back into the loop.

Simple math? Coin flip whether you’ll get the right answer. Ask it to count letters in a word? Apparently that’s PhD-level stuff now. Code generation? Sure, if you enjoy debugging hallucinated APIs that never existed.

The Reddit threads tell the story: “GPT-5 is horrible” with 3,000 upvotes. Users reporting that it’s somehow worse than GPT-4. Plus subscribers discovering their $20/month now buys them 200 messages per week with a model that can’t count to ten reliably.

The Bigger Picture

What really grinds my gears isn’t just the bugs—it’s the arrogance. OpenAI knew this model wasn’t ready. Sam Altman himself warned about “capacity crunches” and “hiccups.” They launched anyway, then had the audacity to remove access to the stable, functional models we’d been using for months.

They’re calling it their “smartest, fastest, most useful model yet.” Meanwhile, users are fleeing to Claude, Gemini, or literally any alternative that doesn’t gaslight them about basic arithmetic.

The benchmarks they’re touting? 94.6% on AIME 2025, 74.9% on SWE-bench—impressive on paper, useless when the model can’t maintain a coherent conversation for more than three exchanges.

The Trust Equation

Here’s what OpenAI doesn’t seem to understand: Trust is earned in drops and lost in buckets. Every forced update, every removed feature, every “we know better than you” decision erodes the relationship with users who’ve been along for the ride since GPT-3.

I used to recommend ChatGPT to everyone. “Try it,” I’d say. “It’ll change how you work.” Now? I’m telling people to steer clear until OpenAI remembers that users aren’t beta testers—we’re customers.

Moving Forward (Without Them)

The ChatGPT app sits on my phone now like a monument to tech hubris. Occasionally I’ll open it, hoping maybe today’s the day they’ve fixed the loops, restored model choice, or just acknowledged they screwed up. Spoiler: they haven’t.

So here I am, deeper in the Claude ecosystem, exploring Gemini for research tasks, and wondering when OpenAI forgot that “Don’t be evil” applies to product decisions too.

GPT-5 wasn’t just a bad launch—it was a masterclass in how to alienate your most loyal users. And unlike their models, that’s something I won’t forget anytime soon.


Update: OpenAI claims they’re doubling rate limits and adding a “reasoning toggle” to fix issues. Too little, too late. When you break trust this badly, patches won’t cut it. We need model choice back, period.