This workaround, however, is slow and labor-intensive, and the use of synthetic data generated by another OpenAI model has led to malfunctions and nonsensical responses.

Initially, GPT-5 was expected to launch in mid-2024, but the ongoing issues have raised doubts about meeting this timeline.

The success of Orion is crucial for OpenAI, as the company's valuation of $157 billion, established in October 2024, heavily relies on the successful development of this next-generation model.

OpenAI's ambitious AI project, GPT-5, also known as Project Orion, is significantly behind schedule and grappling with uncertainties about its effectiveness, as reported by the Wall Street Journal.

The delay in Orion's development has been compounded by a wave of executive departures this year, including notable figures like co-founder Ilya Sutskever and CTO Mira Murati, which may have hindered progress.

These challenges have become evident during extensive training runs, revealing that the quality of training data was less diversified than expected, limiting the model's learning capabilities.

Technical difficulties have plagued the project, with at least two major training runs failing to yield the anticipated results, further complicating the timeline.

Despite undergoing at least two large training runs, the initial run was slower than anticipated, suggesting that future training will be both time-consuming and costly.

The financial implications are substantial, with computing expenses for a six-month training period estimated at around $500 million, raising questions about the project's viability.

While GPT-5 shows some performance improvements over previous models, it has not advanced sufficiently to justify the high operational costs involved in its development.

As OpenAI faces these setbacks, competitors are poised to capitalize on its slowdown, with companies like Amazon increasing their investments in rival firms such as Anthropic.

In response to these challenges, OpenAI is diversifying its data sources by employing personnel to generate new data through coding and mathematical problem-solving.