GPT-4.5, the most expensive AI model ever produced, lacks intelligence and still fails basic math: “Yes, 9.11 > 9.99.”

OpenAI just rolled out GPT-4.5, and the excitement was short-lived. Billed as the most expensive AI model ever produced, it arrived with high expectations, yet it hasn’t delivered anything groundbreaking. The model doesn’t set new benchmarks, introduce revolutionary capabilities, or even fix the simple mistakes that have plagued previous versions. Instead, the big talking point seems to be its ability to generate responses with a “better vibe”—which, depending on who you ask, might not be worth the steep price tag.

OpenAI CEO Sam Altman also took to X to announce the GPT-4.5 launch, highlighting both its strengths and the challenges behind its release.

Calling it “giant, an expensive model,” Altman said,

“Bad news: it is a giant, expensive model. we really wanted to launch it to plus and pro at the same time, but we’ve been growing a lot and are out of GPUs. we will add tens of thousands of GPUs next week and roll it out to the plus tier then. (hundreds of thousands coming soon, and i’m pretty sure y’all will use every one we can rack up.)”

GPT-4.5 is ready!

good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i’ve sat back in my chair and been astonished at getting actually good advice from an AI.

bad news: it is a giant, expensive model. we…

— Sam Altman (@sama) February 27, 2025

Sticker Shock and Limited Access

Even with restricted access, GPT-4.5 carries a jaw-dropping price tag. If you thought Claude’s pricing at $15 per million tokens was steep, OpenAI has taken things to another level. GPT-4.5 charges $75 per million input tokens and $150 per million output tokens—making it the most expensive model to date. Right now, access is limited to OpenAI’s $200-per-month Pro users, meaning most casual users won’t get to try it anytime soon.

Source: OpenAI

The promise? A model that sounds more human-like in conversation. But does that justify the cost?

GPT-4.5 Lacks Intelligence And Still Makes The Same Mistakes

It’s hard to ignore the irony. A model costing hundreds of millions to develop—and requiring users to shell out serious cash—still can’t pass elementary school math. While OpenAI claims it hallucinates less, examples like this raise questions about how much real progress has been made.

Despite all the hype, GPT-4.5 still fumbles basic logic and math. When asked whether 9.11 is greater than 9.99, it confidently responded: “Yes, 9.11 > 9.99.”

An X user, Deedy, put GPT-4.5 to the test with a simple math question: 9.11>9.99? The model got it wrong five times in a row.

“Tried this 5 times on ChatGPT 4.5 and it gets it wrong every time,” Deedy wrote on X, sharing the baffling mistake.

Tried this 5 times on ChatGPT 4.5 and it gets it wrong every time. pic.twitter.com/3aH6ajIpF7

— Deedy (@deedydas) February 28, 2025

On coding benchmarks, things don’t look much better. The AER Polyglot Coding Benchmark shows that GPT-4.5 performs worse than DeepSeek, a model that outperforms other top AI models and costs significantly less to run. For a company trying to justify sky-high AI development costs, this isn’t a great look.

Sam Altman on GPT-4.5: The Good and the Bad

Altman called GPT-4.5 the first AI model that “feels like talking to a thoughtful person,” adding that he’s had moments where its responses left him genuinely surprised by their quality.

But there’s a catch. He acknowledged that GPT-4.5 is a massive and costly model, with GPU shortages preventing OpenAI from rolling it out to all users immediately. While Pro users have access now, Plus subscribers will have to wait until OpenAI adds tens of thousands of GPUs next week—with hundreds of thousands more on the way.

Not Built to Dominate Benchmarks

Altman also set expectations, making it clear that GPT-4.5 isn’t designed to excel at reasoning or set new benchmark records. Instead, he described it as a “different kind of intelligence” with a unique quality that he finds compelling.

The rollout issues highlight a broader challenge for OpenAI: keeping up with demand while pushing forward with increasingly resource-intensive models. Altman admitted that predicting GPU needs has been tricky, but he expects OpenAI to scale up quickly.

For now, the focus is on making GPT-4.5 available to more users—and seeing if its “thoughtful” responses justify the cost.

Sam Altman Didn’t Even Show Up

For a launch this big, you’d expect OpenAI’s CEO, Sam Altman, to be front and center. Instead, he skipped the event, reportedly staying with his newborn while interns handled the demo. It’s hard not to compare this with 2023, when Altman and other AI leaders were calling for a pause on advanced AI development over concerns about existential risks.

The same people who once urged governments to regulate these models are now rolling out the most expensive one yet—with no major breakthroughs to show for it.

taking care of my kid in the hospital.

the team has got this!

— Sam Altman (@sama) February 27, 2025

Elon’s Grok Takes the Lead?

Right now, AI enthusiasts and industry insiders seem to be shifting their focus to xAI’s Grok, with betting markets calling it the best AI model currently available. That doesn’t mean OpenAI is out of the race—GPT-5 is expected to arrive later this year, and many still believe it will eventually take the lead. But the odds aren’t looking as strong as they once did.

For OpenAI, this could pose a real problem. The company has raised billions of dollars, transitioned into a for-profit structure, and is banking on continued investor confidence. Altman remains bullish on scaling AI models indefinitely, but that depends on whether they can justify the investment.

The Road Ahead

OpenAI claims GPT-5 will be a “router” model that selects the best AI for each prompt rather than a single model doing all the work. That could mean shifting expectations—away from bigger models and toward better specialization.

But for those expecting a major leap forward, the rollout of GPT-4.5 has been a letdown. Some were hoping for a future with superintelligent AI, but instead, we’re debating whether an overpriced chatbot gives off the right “vibes.”

For developers and engineers, this might not be a bad thing. AI coding assistants are useful, but they still require human oversight—and that’s unlikely to change anytime soon.

So, while some feared we’d all be fighting robots by now, it turns out we’re still just arguing about bad math and expensive subscriptions.

Here’s a video of the OpenAI team discussing GPT-4.5 during the demo.

Today we’re releasing a research preview of GPT-4.5—our largest and best model for chat yet.

Rolling out now to all ChatGPT Pro users, followed by Plus and Team users next week, then Enterprise and Edu users the following week. pic.twitter.com/br5win5OEB

— OpenAI (@OpenAI) February 27, 2025