OpenAI’s latest model, GPT-4.5, has landed with a resounding thud. Despite the company’s attempt to market it as “the largest and most knowledgeable model yet,” the broader response has been far from enthusiastic. Critics and AI experts alike are calling it overpriced, underwhelming, and, most damningly, not particularly useful. Even OpenAI seems reluctant to fully back its own creation, avoiding the term “frontier model” and instead focusing on its supposed improvements in emotional intelligence. The problem? That’s not what people want in an advanced AI model.
More details on the situation can be found here, but the general sentiment is clear: GPT-4.5 doesn’t justify its astronomical cost. At $75 per million input tokens and $150 per million output tokens, it is 30 times more expensive than GPT-4o. One might assume such a steep price increase would come with a leap in capabilities, but experts suggest otherwise. According to an anonymous AI researcher quoted by Ars Technica, the model is essentially a “lemon.” Slower than its predecessors and still prone to hallucination, it appears to be little more than a pricier, more verbose chatbot.
That last point is crucial. OpenAI’s messaging around GPT-4.5 leans heavily on its ability to “engage in warm, intuitive, natural, flowing conversations.” It’s designed to be emotionally intelligent, to pick up on subtle nuances in human speech, and to respond with greater sensitivity. But at its core, a large language model is supposed to be an analytical tool, not a digital therapist. The backlash has been swift, with users joking that GPT-4.5 is more suited for venting frustrations than solving complex problems. One critic summed it up perfectly: “I want a smart model, not a model that understands my feelings.”
Even OpenAI’s internal testing results don’t inspire confidence. The MIT Technology Review reports that GPT-4.5 still fabricates information at an alarming rate, making up facts 37% of the time in key tests. That’s a stunning failure rate for a system that’s supposed to be more reliable and knowledgeable than its predecessors. If a supposed upgrade still hallucinates nearly four out of ten times, what exactly justifies the higher price?
Mark Chen, OpenAI’s chief research officer, made an even more baffling admission—stating that the benchmarks used to evaluate GPT-4.5’s progress are “vibes-based.” This is an astonishing thing to hear from a company that claims to be at the forefront of AI research. “Vibes” are great for music recommendations, but they don’t belong anywhere near AI development. The idea that OpenAI is shaping its flagship model based on subjective feelings rather than measurable, scientific benchmarks is a clear indication of just how lost the company has become.
OpenAI has enjoyed years of dominance in the AI space, but missteps like this raise serious questions about its direction. The company once positioned itself as a leader in cutting-edge AI, pushing the limits of what language models could achieve. Now, it seems more concerned with making its models sound friendly while charging exorbitant rates for minimal improvements. If this is the future of OpenAI’s product line, it may not remain the industry leader for long.