OpenAI has released its newest AI model, GPT-4.5, which promises improved performance but comes at a significant cost. The new model is 30 times more expensive for input and 15 times more expensive for output compared to its predecessor, GPT-4o. While it provides marginally better results, experts are questioning the value of this investment.
Gary Marcus, an outspoken critic of OpenAI, called the release “a nothing burger” due to the high price tag. Another expert described it as a “lemon,” citing the significant increase in cost without proportional improvements. Even Andrej Karpathy, a former researcher at OpenAI, praised GPT-4.5 for its subtle yet impressive advancements, but noted they are not easily quantifiable.
OpenAI acknowledged these limitations and framed the release as a “Research Preview” to avoid disappointing users. The company emphasized that GPT-4.5 is not a replacement for GPT-4o, but rather an addition to their API offerings. However, benchmark results show that GPT-4.5 falls short of OpenAI’s simulated reasoning models in various tests, scoring significantly lower on math competitions and science assessments.
The high cost of GPT-4.5 is further evident in the significant difference in input processing costs compared to other models. This raises questions about whether this investment will yield long-term benefits or if it represents a step backward due to diminishing returns in training large language models.
Source: https://arstechnica.com/ai/2025/02/its-a-lemon-openais-largest-ai-model-ever-arrives-to-mixed-reviews