OpenAI’s GPT-5 reportedly falling short of expectations

OpenAI's development of GPT-5, codenamed "Orion," is reportedly behind schedule and facing challenges, according to The Wall Street Journal. Despite over 18 months of work and at least two large training runs, the results have not yet justified the significant costs involved.

An initial training run was slower than expected, and while GPT-5 shows improvements over previous models, its performance advancements are not substantial enough to warrant the high operational costs. OpenAI has been diversifying its data sources, hiring people to create fresh data (e.g., writing code or solving math problems) and using synthetic data generated by another model, o1.

OpenAI has confirmed it will not release Orion this year and has not commented further on the reports.

Read the full article