OpenAI unveiled GPT-4.5 on Thursday, marking a new competitive phase in the artificial intelligence race as companies continue to roll out increasingly capable – and increasingly expensive – models. This release comes just one day after Anthropic launched Claude 3.7 Sonnet and a mere week after xAI’s introduction of Grok-3 and DeepSeek’s hint at a forthcoming model.
The new model from OpenAI, however, is not for the faint of heart in terms of cost. With an API price tag of $75 per million input tokens and $150 per million output tokens, GPT-4.5 is ten times pricier than Claude 3.7 Sonnet – a potentially prohibitive cost for many developers and start-ups wishing to build on the technology. For comparison, its predecessor, GPT-4o, cost $2.50 per million tokens of input and $10.00 per million tokens of output, making GPT-4.5 a whopping 2900% more expensive for inputs and 1300% pricier for outputs.
OpenAI’s CEO, Sam Altman, candidly acknowledged the model’s significant resource requirements in his announcement, stating, “Bad news: It is a giant, expensive model.” But he also pointed out a unique selling point: “There’s a magic to it I haven’t felt before.”
Interestingly, the high cost is not justified by an increase in intelligence, but rather by a more human-like, nuanced interaction. OpenAI highlights the model’s “vibes,” or its emotional intelligence, warmth, and collaborative feel. GPT-4.5 reportedly outperformed other models in a “Vibes test set” measuring creative intelligence and conversational quality.
The product demonstrations, while not introducing anything groundbreaking, did show some improved nuance in GPT-4.5. The difference was primarily in tone. For example, when asked why the sea water is salty, GPT-4.5 responded in simpler terms compared to previous models.
In terms of performance, GPT-4.5 paints a mixed picture. It scores 71.4% on GPQA, a science evaluation, compared to GPT-4o’s 53.6%. However, it still lags behind OpenAI’s o3-mini model, which scores 79.7% through its reasoning capabilities.
Altman described the model as “the first model that feels like talking to a thoughtful person,” adding, “I have had several moments where I’ve sat back in my chair and been astonished at getting actually good advice from an AI.”
GPT-4.5 represents OpenAI’s latest efforts to scale up unsupervised learning, increasing world knowledge, intuition, and reducing hallucinations. The model’s development necessitated massive technical innovation, including the construction of new inference systems to serve such a large model efficiently, the use of low-precision training to maximize GPU usage, and training across multiple data centers simultaneously.
As consumer expectations for AI continue to rise and competition in the space intensifies, only time will tell if GPT-4.5’s “different kind of intelligence” and enhanced “vibes” justify its substantial resource requirements and steep pricing. GPT-4.5 is currently available for Pro users who pay $200 per month, with Plus users paying $20 a month gaining access to the model next week.
