Anthropic deletes AI security commitments in the Biden era, and policy trends attract attention
542
English
English
Recently, the technology media The Decoder released a report on OpenAI's latest model GPT-4.5, which aroused the industry's doubts about the cost-effectiveness of the model. According to official data, although GPT-4.5 has improved its performance, its cost has increased significantly compared to the previous version of GPT-4o.
Specifically, the proportion of GPT-4.5 performing better than GPT-4o in multiple aspects is: 63.2% of professional inquiry, 57% of daily inquiry, and 56.8% of creative tasks. However, these improvements are only between 6.8% and 13.2%. What's more remarkable is that in terms of usage costs, the cost of inputting the token is as high as $75 per million, while the cost of outputting the token is $150. By comparison, the GPT-4o costs only $2.50 for input and $10 for output. This means that the input cost of GPT-4.5 is 30 times that of GPT-4o, and the output cost is 15 times.
OpenAI said that GPT-4.5 will not replace GPT-4o, and is currently evaluating the feasibility of providing the model through the long-term API. OpenAI is also actively collecting user feedback to determine whether GPT-4.5 can show enough unique value to support its high operating costs. It is worth noting that if GPT-4.5 ultimately decides to stop the service, then application developers who rely on the model for development may face some dilemma.
In response to the current challenges, OpenAI may launch an optimized version of GPT-4.5 Turbo to reduce usage costs. However, given the limited performance improvement of GPT-4.5, it remains questionable whether it can bring significant advantages. Although the GPT-4Turbo in the past has been optimized in terms of speed and price, its quality is generally not as good as the original GPT-4.
Against the backdrop of rapid development of AI technology, OpenAI faces major challenges in how to balance model performance and usage costs. In the future, whether a reasonable path can be found will directly affect its market competitiveness and user acceptance.
