OpenAI’s new GPT-5 model does not seem to meet expectations.
OpenAI’s long-awaited GPT-5 model, successor to GPT-4 codenamed Orion, does not appear to be living up to high expectations. According to a report by The Wall Street Journal, despite huge investments and 18 months of work, the development has not yet produced the desired progress that justifies the high cost.
Slow development
According to the report, the project faces delays and obstacles. Two rounds of training have produced less impressive results than hoped. The first round of training was slower than planned, raising doubts about the time needed for the next steps. Although GPT-5 is reportedly performing better compared to earlier versions, it has not yet progressed far enough to justify its cost.
To develop the model, OpenAI did more than just use public data and licensing agreements. The company hired people to generate original data, such as programming tasks and mathematical solutions, and also used synthetic data produced by an earlier model, o1.