AI enthusiasts and skeptics alike had their eyes on OpenAI’s latest release, GPT-4.5. But instead of a groundbreaking leap, we got a model that’s more expensive than ever, with little to show for it in terms of actual performance. Is this the beginning of the end for AI hype, or just a temporary plateau?
The Reality of GPT-4.5
On February 28, 2025, OpenAI unveiled GPT-4.5, the most expensive AI model ever created. But despite the hype, it failed to deliver any earth-shattering advancements. Here’s what we know:
- Cost: GPT-4.5 is five times more expensive than competitors like Claude. It costs $75 per million input tokens and $150 per million output tokens.
- Access: Currently, it’s only available to $200/month Pro users.
- Performance: OpenAI claims fewer hallucinations, but the model still struggles with accuracy, even making basic spelling mistakes.
The Vibes Benchmark
OpenAI’s biggest selling point for GPT-4.5? Vibes. According to their launch event, the model is designed to chat more naturally. They even introduced a Vibes Benchmark to measure creative thinking. But how do you quantify “good vibes”? That’s highly subjective.
While the model feels more conversational, it doesn’t necessarily outperform competitors in raw intelligence or reasoning. It still fails at certain factual tasks and isn’t self-aware—when asked, it didn’t even recognize itself as GPT-4.5 and instead claimed its training cutoff was October 2023.
Programming and Science: Still a Letdown
If you’re a developer or data scientist hoping for a breakthrough, prepare for disappointment. GPT-4.5 underperforms on coding benchmarks like the AER Polyglot Coding Benchmark, lagging behind competitors like DeepSeek. Worse yet, it’s also hundreds of times more expensive.
For those keeping score, xAI’s Grok is currently considered the best AI model, at least according to betting markets. OpenAI, on the other hand, is losing ground, despite its multi-billion-dollar ambitions.
The Future of AI: A Plateau, Not a Singularity?
A few years ago, tech leaders were panicking over the potential dangers of advanced AI. Now? We might be hitting a limit. OpenAI’s Sam Altman insists there’s “no wall” and that they can keep scaling these models indefinitely. But there’s speculation that GPT-5 will be more of a router—picking the best model for your query—rather than a true upgrade in intelligence.
For now, we’re living in a strange dystopia where superintelligent AI hasn’t arrived, and nothing revolutionary is happening. If you’re a programmer, though, this might be good news—AI tools can assist coders, but they’re nowhere near replacing them. If anything, now’s a great time to double down on learning programming.