Close Menu
Tech

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

userBy No Comments6 Mins Read

Openai has just launched its GPT-4.5 and the excitement was short-lived. It was billed as the most expensive AI model ever produced and arrived with high expectations, but it offers nothing groundbreaking. This model does not set new benchmarks, introduce innovative features, or correct the simple mistakes that bothered previous versions. Instead, the big topic appears to be the ability to generate responses with a “better atmosphere.”

Openai CEO Sam Altman also used X to announce the release of the GPT-4.5, highlighting both its strengths and the challenges behind its release.

He calls it “a Giant, an expensive model,” Altman said.

“Bad news: It’s a huge, expensive model. We really wanted to boot into the plus and pro at the same time, but we’re continuing to grow a lot and coming out of the GPU. Next week we’ll add tens of thousands of GPUs and roll out into the plus layer. (There will be hundreds of thousands coming soon, and I’m sure you’ll use everything we can get.)

Sticker shock and access are restricted

Even with limited access, the GPT-4.5 comes with a price tag to drop your chin. If you think Claude’s pricing is steep with a token of $15 per million, Openai has taken things to another level. GPT-4.5 charges $75 per input token and $150 per million output token. Currently, access is limited to Openai’s $200 per month professional users. This means that most casual users won’t be able to try it out right away.

Source: Openai

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

GPT-4.5 has no intelligence, and still makes the same mistake

It’s hard to ignore the irony. Models that cost hundreds of millions of dollars to develop and require users to pay serious cash will not be able to pass elementary school maths. Openai claims there are fewer hallucinations, but examples like this raise questions about how much progress has actually been.

Despite all the hype, GPT-4.5 still shakes basic logic and mathematics. When asked whether 9/11 exceeded 9/99, it replied with confidence: “Yes, 9/11>9/99”.

X user, Deedy tested GPT-4.5 with a simple math question: 9.11> 9.99? The model was wrong five times in a row.

“I tried this five times on ChatGpt 4.5 and it’s wrong every time,” Deedy wrote to X, sharing the inexplicable mistake.

Things don’t look that good in benchmark coding. The AER polyglot coding benchmark indicates that GPT-4.5 is worse than DeepSeek. This is a model that surpasses other top AI models and is significantly less costly to run. For businesses looking to justify the development costs of Sky-High AI, this isn’t a great look.

Sam Altman in GPT-4.5: Good and Evil

Altman called the GPT-4.5 the first AI model “feels like talking to thoughtful people,” adding that there were moments when the response really surprised him at his quality.

But there’s a catch. He acknowledged that GPT-4.5 is a large and expensive model, and that the GPU shortage will prevent Openai from deploying it to all users immediately. Professional users are currently visiting, but subscribers will have to wait for hundreds of thousands of GPUs next week, and hundreds of thousands more GPUs.

Not built to control the benchmark

Altman also sets expectations and makes it clear that GPT-4.5 is not designed to excel inference or set new benchmark records. Instead, he described it as “a different kind of intelligence” with a compelling unique quality.

The rollout issue highlights a broader challenge for Openai. Keep up with demand as we advance more and more resource-intensive models. Altman admitted that it’s difficult to predict GPU needs, but hopes Openai will scale up quickly.

For now, we’re focusing on making GPT-4.5 available to more users and seeing if its “thoughtful” response justifies the cost.

Sam Altman did not appear

For this big launch, you’ll hope that Openai CEO Sam Altman will be front and center. Instead, he skipped the event and reportedly stayed with the newborn while the intern handled the demonstration. It’s difficult to compare this to 2023, when Altman and other AI leaders were calling for a pause on advanced AI development over existential risk concerns.

The same people who once urged the government to regulate these models are now rolling out the most expensive models.

Will Elon’s Grok lead?

Currently, AI enthusiasts and industry insiders seem to have shifted their focus to Xai’s Grok, and the betting market is called the best AI model available today. That doesn’t mean Openai is out of the race. The GPT-5 is expected to arrive later this year, and many believe they will ultimately lead. However, the odds don’t look as strong as they used to be.

For Openai, this can cause real problems. The company has raised billions of dollars, moved into a for-profit structure and banking the bank to ongoing investor trust. Altman remains bullish by scaling AI models indefinitely, but that depends on whether they can justify their investment.

The road ahead

Openai claims that the GPT-5 is not a single model that does all the work, but rather a “router” model that selects the best AI for each prompt. That could mean changing expectations for better specialization from a larger model.

But for those hoping to move forward, the GPT-4.5 rollout was a disappointment. Some wanted a future with close AI, but instead they are debating whether a high-value chatbot will release the correct “vibe.”

For developers and engineers, this may not be a bad thing. AI coding assistants are useful, but still require human surveillance. It’s unlikely to change it anytime soon.

So, while some people are afraid that we all fight robots, we find ourselves still discussing bad maths and expensive subscriptions.

Here is a video from the Openai team discussing GPT-4.5 during the demonstration.




Source link

Share.

Related Posts

Add A Comment
Leave A Reply