OpenAI’s GPT-5, code-named Orion, faces delays and performance concerns amid rising costs.

OpenAI’s highly anticipated AI model, GPT-5, appears to be facing significant hurdles, according to a new report from The Wall Street Journal. Development of the model, code-named Orion, has fallen behind schedule, with results that, as of now, fail to justify the massive costs involved.

The delays are not entirely unexpected, as an earlier report from The Information suggested that GPT-5 may not represent the substantial leap forward many had anticipated. However, the new report sheds light on the challenges OpenAI has encountered during the 18-month development process. Despite completing at least two large training runs, the company’s progress has been slower than anticipated, particularly during the initial phases. The complexity and cost of continuing to train GPT-5 are mounting, and while the model is reportedly more powerful than its predecessors, the advancements do not yet justify the hefty price tag attached to keeping the model operational.

One of the key hurdles OpenAI faces is the high computational cost of running such a large-scale project. Although GPT-5 has shown improvements in performance compared to previous iterations, it appears to be far less revolutionary than initially hoped. OpenAI has also reportedly had to get creative with its data sourcing in an attempt to fuel the model’s growth. Instead of relying solely on publicly available datasets or licensing deals, the company has hired additional personnel to create new data through tasks such as coding or solving mathematical problems. Furthermore, OpenAI is utilising synthetic data generated by another model in its arsenal, o1, to help fill in gaps and accelerate GPT-5’s training.

Despite these efforts, the model’s progress remains inconsistent, and it has not yet reached a stage where OpenAI can justify the ongoing costs of development and operational maintenance. The Wall Street Journal report points to the need for new strategies, as GPT-5’s slow and costly development raises questions about whether it can meet the lofty expectations set for it.

While OpenAI did not immediately respond to requests for comment, it has previously stated that it would not release a model named Orion this year, indicating that the company is well aware of the delays and difficulties surrounding the project. As OpenAI pushes forward with GPT-5, many are left wondering whether the AI community will see the next evolution of language models sooner rather than later—or if the model will need further revisions and recalibrations before it can make the expected leap.