The Wall Street Journal recently reported that OpenAI’s progression towards its next significant model, GPT-5, is experiencing delays. The results thus far do not appear to justify the substantial costs associated with the project.

This report aligns with a previous one by The Information, hinting that OpenAI is considering alternative strategies, as GPT-5 may not provide as significant an advancement as its predecessors. The Wall Street Journal article shares more insights into the 18-month journey of GPT-5’s development, codenamed Orion.

OpenAI has allegedly undertaken at least two significant training runs aimed at enhancing the model by subjecting it to vast amounts of data. The first such run progressed slower than anticipated, indicating that a larger run could be both time-intensive and expensive. Although GPT-5 reportedly outperforms its predecessors, its advancement does not yet warrant the cost of sustaining the model’s operation.

In addition to using publicly accessible data and licensing agreements, the Wall Street Journal reports that OpenAI has recruited individuals to generate new data by coding or solving mathematical problems. Furthermore, OpenAI is leveraging synthetic data produced by another model, o1.

OpenAI has not provided an immediate response to these claims. However, the company has previously stated that it would not be launching a model codenamed Orion within the current year.