On February 27, 2025, OpenAI officially released its latest chat model GPT-4.5, and the news quickly sparked widespread discussion on Twitter (now the X platform). As OpenAI's most powerful conversation model to date, GPT-4.5 has achieved a significant breakthrough in performance, but its high price has also caused users and developers to talk about it. At the same time, industry insiders have had a fierce debate over the comparison of its capabilities with existing models.
The pricing of GPT-4.5 is 13 to 30 times higher than the previous GPT-4o. Specifically, its API price is $75 per million input token and $150 per million output token. Netizen @verysmallwoods jokingly said: "After one meal, you can have a hot pot meal in one Q&A!" In contrast, the input price of GPT-4o is US$5 per million token and the output price is US$15, while domestic competitors such as DeepSeek are still seizing market share through "off-peak price reduction". Such a significant price increase has made many users call it "outrageous". Some developers even said that OpenAI may need to reevaluate whether to continue to provide this version of the API.
Netizen @AlchainHust even revealed that when he entered the price information of GPT-4.5 into GPT-4o and asked his opinion, GPT-4o actually said "it's impossible to be true, it's too outrageous" and questioned the source of the data. This interaction has aroused the jokes of netizens, and also highlighted the controversy of GPT-4.5 pricing.
Despite the staggering price, the performance of GPT-4.5 is still the focus of attention. GPT-4.5 significantly improves intuitive understanding and knowledge breadth of the world by extending unsupervised learning and reduces the common "illusion" problems in the model. Unlike the O1 series that relies on inference chains, GPT-4.5 relies more on large-scale data and computing power to optimize performance. In SimpleQA benchmark, the accuracy of GPT-4.5 reached 62.5%, and the hallucination rate dropped to 37.1%, better than the performance of GPT-4o, o1 and o3-mini. This advancement is considered a manifestation of its higher "emotional intelligence", especially when dealing with complex dialogue and open-minded issues.
However, @AlchainHust pointed out that despite the performance improvement of GPT-4.5, it seems to be unbiased on the multiple model evaluation benchmarks, which are not proportional to its high price. In contrast, competitors such as Anthropic's Claude 3.5 Sonnet (which is said to be priced at 25 times the price of GPT-4.5) have a better advantage in terms of cost-effectiveness.
OpenAI positioned GPT-4.5 as a model that "pulls intelligence to the extreme through scale-up pre-training". According to Twitter users, this strategy has been verified by previous predictions by Ilya Sutskever, former chief scientist of OpenAI, that is, by increasing data and computing power, the model performance curve can continue to rise. Currently, GPT-4.5 is open to Pro users, while Plus and Team users will have to wait until next week to use it. However, the high costs also cause internal doubts about its commercialization prospects, and may even affect subsequent API supply plans.
Judging from the user feedback from the X platform, the release of GPT-4.5 has brought both surprises and questions. On the one hand, its improvement in understanding and dialogue quality is considered a sign that AI technology has reached a new height; on the other hand, the soaring price has discouraged many small and medium-sized developers and ordinary users. At the same time, the low-price strategies of domestic manufacturers such as DeepSeek are intensifying market competition, and it is still unknown whether OpenAI's high-end route can continue to be recognized.
In general, the debut of GPT-4.5 has undoubtedly injected new topics into the AI field. Is its ability improvement enough to support its high pricing? How big is the gap between it and GPT-4o, o1 and even other manufacturer models? These questions may only be concluded after more user experience and review data are released. In this game of "performance and price", OpenAI's next chess game is worth paying close attention to.