Reddit revolt: 3,000 upvotes on "GPT-5 is horrible" thread, 3,000+ signed petition for old models. Markets crashed: OpenAI odds dropped from 75% to 12%, Google shot up to 80%. Sam Altman admitted they "totally screwed up" the launch. Developers loved it, regular users hated it - OpenAI optimized for the wrong audience. Router system was broken, Plus subscribers hit with new limits without warning.
So OpenAI thought they were dropping a mic on August 7th. Sam Altman went live, hyping GPT-5 as their "smartest, fastest, most useful model yet." The whole thing felt like a victory lap - breakthrough benchmarks, AGI vibes, the works.
Two weeks later? The internet is absolutely roasting them.
When the Numbers Are This Brutal
The data from Reddit tells a devastating story.
I dove into over 10,000 Reddit discussions from the week after launch, and the data is... yikes.
A thread called "GPT-5 is horrible" hit nearly 3,000 upvotes and 1,200 comments. Multiple threads begging for GPT-4o's return got thousands of upvotes. Over 3,000 people actually signed a petition demanding access to the old models. A petition! When's the last time you saw people petition to get an older version of software back?
Market Reaction
On Polymarket, OpenAI's odds of having the best AI model cratered from 75% to 12% within hours. Google's odds shot up to 80% in the same timeframe. Some day trader made $10,000 in a few hours just betting against GPT-5's popularity.
When Sam Altman Admits He Screwed Up
A rare moment of Silicon Valley honesty.
Here's something you don't see every day in Silicon Valley: a CEO actually admitting they messed up. At a press dinner, Sam Altman straight-up said they "totally screwed up" the GPT-5 launch.
GPT-5's personality felt like talking to a burnt-out corporate drone instead of the chatty, helpful AI people had grown to love. Users started describing it as an "overworked secretary."
OpenAI scrambled to fix things - brought back GPT-4o as an option and promised to make GPT-5 less robotic. But the damage was already done.
What Users Were Actually Dealing With
The real-world experience was far from the benchmark hype.
Reddit user RunYouWolves summed it up perfectly: "It's like my chatGPT suffered a severe brain injury and forgot how to read. It is atrocious now."
People were getting shorter, less helpful responses. Creative writing tasks that used to work great suddenly became a struggle. Then the real bombshell: their fancy new router system was broken. The thing that was supposed to automatically pick between fast and smart modes? Yeah, that was "out of commission for a chunk of the day."
The Plus Subscriber Problem
ChatGPT Plus subscribers - people paying $20 a month - suddenly found themselves with new limitations. Only 200 messages per week with GPT-5. No access to the models they actually preferred. Zero warning about any of this.
Need help implementing this?
50+ implementations · 60% faster · 2-4 weeks
Plot Twist: Developers Actually Liked It
A tale of two audiences with very different needs.
While regular users were losing their minds, developers were actually pretty happy. Cursor called GPT-5 "the smartest model they've used." Vercel said it was "the best frontend AI model" they'd tested.
| Audience | Reaction | Why |
|---|---|---|
| Regular Users | Hated it | Lost personality, creative flow, conversational feel |
| Developers | Loved it | Better coding, debugging, complex technical tasks |
| Plus Subscribers | Furious | New limits, lost model access, no warning |
OpenAI optimized for the wrong audience. They made GPT-5 great for professional work but forgot that most people just want interesting conversations and help with creative projects.
When Benchmarks Don't Tell the Whole Story
Impressive numbers that missed the point entirely.
On paper, GPT-5 looked incredible: 94.6% on math competitions, 74.9% on coding benchmarks, 45% fewer hallucinations, 80% fewer reasoning errors.
But one Reddit comment nailed the problem: "I like how in the demo they were like 'if it gets something wrong, no worries, just ask again.' How is that better?"
Benchmarks measure specific tasks under controlled conditions. They don't capture whether an AI feels natural to talk to or whether people actually enjoy using it.
Security Issues Too
As if the backlash wasn't enough.
As if the user backlash wasn't enough, security researchers at Adversa found a significant vulnerability. Users could game the routing system with specific phrases to get responses from older, potentially less safe models.
So not only was the new system broken and unpopular, it was also less secure than what came before.
Timeline of Chaos
Eleven days from triumph to disaster.
How It All Went Down
- 1August 7: OpenAI launches GPT-5 with all the fanfare
- 2August 8: Reddit starts absolutely losing it
- 3August 9: Prediction markets flip to Google
- 4August 10: Sam Altman caves and brings back GPT-4o
- 5August 11: Damage control AMA ('please don't leave us')
- 6August 18: Altman finally admits they 'totally screwed up'
Eleven days from hero to zero. That's got to be some kind of record.
Terrible Timing for OpenAI
A PR disaster during critical fundraising.
This disaster is unfolding right as OpenAI is supposedly trying to raise money at a $500 billion valuation. That's a tough sell when your latest product is getting roasted by the entire internet.
Meanwhile, competitors are loving this. Google's Gemini started looking better in prediction markets. Claude keeps getting mentioned as the go-to alternative. Meta and everyone else are probably popping champagne.
What We Learned
Critical takeaways for the AI industry.
A few things became crystal clear:
User feedback actually matters now. The speed and intensity of the backlash forced OpenAI to make changes within days.
AI personality is way more important than anyone realized. People got genuinely emotional about losing GPT-4o's conversational style. These AI tools are becoming less like software and more like relationships.
The "rip off the bandaid" approach doesn't work when people are already attached to what they have. You can't just replace everyone's favorite AI without warning.
Even market leaders can stumble badly with poor execution. Being first doesn't mean you get to stay first.
The Bottom Line
What GPT-5's rocky launch tells us about the future of AI.
This whole fiasco perfectly captures what's happening in AI right now. OpenAI was celebrating benchmark scores while users were mourning the loss of an AI personality they'd grown attached to.
The companies that figure out how to keep advancing technically without losing that human touch are going to win big. Because here's what GPT-5 taught us: sometimes the "best" product on paper isn't the one people actually want to use.
Keep Reading
Want to Navigate the AI Landscape?
We help businesses cut through the hype and implement AI tools that actually work for their specific needs. Get a free consultation to explore your options.
Get Free Strategy Call15 min • No commitment • We'll send you a customized roadmap
“They helped us deploy an AI chatbot in 2 weeks that would have taken us 3 months internally.”
— Startup Founder, SaaS Company