😈GPT-5: Love it or hate it?

Well, well, well... OpenAI just dropped GPT-5 (August 7th), and let me tell you, the internet has feelings about it.

Sam Altman and crew rolled out the red carpet with a live-stream that promised "PhD-level expertise for everyone,". Also, he tried live tweeting while the launch

Which honestly sounds like something you'd hear at a late-night infomercial, but here we are living in the future, folks.

The Good, The Bad, and The Mislabeled Charts

First off, let's talk about that presentation.

OpenAI clearly didn't get the memo about Apple's keynote magic because this thing was... rough around the edges.

We're talking mislabeled charts (classic!), physics demos that gave wrong answers until they frantically re-ran the prompts, and speakers looking off-camera like they were reading cue cards their dog had chewed up.

The live coding demo even failed to hot-reload at one point, which honestly made it feel more authentic.

Nothing says "real software development" like things breaking when you need them most, am I right?

But here's where it gets interesting, despite the janky presentation, the actual tech seems pretty solid.

What's Actually New (And Why Your Wallet Might Be Happy)

GPT-5 isn't just one model anymore, it's like having a smart intern who knows when to phone a friend.

The system automatically routes between "gpt-5-main" (the speedy one) and "gpt-5-thinking" (the heavy lifter) depending on what you're asking.

No more toggling between models like you're playing some kind of AI slot machine.

The performance numbers are genuinely impressive:

• Coding benchmarks jumped to 74.9% on SWE-bench (that's the hard stuff)

• Medical reasoning went from basically 0% to 46.2% on their new HealthBench

• 45% fewer hallucinations than GPT-4o (finally!)

And developers are probably doing happy dances because API pricing dropped to $1.25 per million input tokens – that's Gemini territory, folks.

Plot Twist: The Internet Is... Divided?

Here's where things get spicy.

While enterprise customers and tech reviewers are singing praises, Reddit and Twitter are having what can only be described as a collective meltdown.

The main complaints?

OpenAI basically yeeted all the beloved older models (goodbye GPT-4o, we hardly knew ye) and replaced them with this router system that some users feel gives shorter, more generic answers.

To read more, click here

Plus subscribers went from unlimited access to some models to a capped 200 messages per week with the thinking model.

One Reddit thread titled "GPT-5 is horrible" hit 3K upvotes faster than you can say "shrinkflation." Ouch.

The Real Tea

Look, what's really happening here is OpenAI made a classic tech company move:

They optimised for cost and efficiency while marketing it as a capability upgrade. And honestly, that's probably smart business, even if it ruffled some feathers.

The unified model approach is actually pretty clever, it's like having a team of specialists who automatically hand off work to whoever's best suited for the job.

This is probably where AI is heading anyway: orchestrated systems of specialised models working together.

The Bottom Line

GPT-5 isn't AGI (sorry, sci-fi fans), but it's a solid step forward in making AI more reliable and accessible.

The enterprise folks are happy, developers are saving money, and everyday users are... well, they're adjusting.

Will OpenAI smooth things over by bringing back model choice and fixing the router quirks?

Probably. Will this change the trajectory of AI development? Definitely.

And hey, at least those 1,000+ OpenAI employees who stuck around instead of jumping ship to Meta got bonuses ranging from $200K to $1.5M.

Not a bad reward for weathering the storm and helping push out what might be the most scrutinised AI launch in history.

The AI wars continue, and we're all just along for this wild, mislabeled-chart-filled ride.

P.S. – If you want to see how GPT-5 actually performs, try asking it to improve your prompts.

Apparently, that's one of OpenAI's recommended use cases now.

Meta-prompting: because why think for yourself when AI can think about your thinking?

Before You Go

Using GPT-5? We want to hear about it!

Send us your story for a chance to be in our next newsletter.

New to AI but curious about what's possible?

Subscribe here for weekly tutorials that actually make sense.

No jargon, no hype, just step-by-step guides you can follow.