OpenAI’s next-generation language model, GPT-5, has run into technical and cost headwinds that are slowing its development, according to an Aug. 1 investigation by The Information. People familiar with internal testing told the publication that the new system shows stronger performance in coding, mathematics and multi-step tasks, yet the improvements are incremental compared with the leap from GPT-3 to GPT-4. Engineers are banking on advances in reinforcement learning to narrow the gap, the report said, while outside testers cited in separate posts have clocked GPT-5 at 45 percent on a demanding lateral-reasoning benchmark—outperforming leading competitors. Despite the remaining hurdles, OpenAI is still aiming to introduce the model in early August, with speculation centring on an Aug. 5 debut. The tempered progress highlights broader challenges facing the sector. Meta Platforms disclosed plans to offload roughly $2 billion in data-center assets to outside partners to contain ballooning infrastructure bills, and industry-wide capital expenditure on artificial-intelligence projects has already topped $155 billion this year, according to recent Guardian estimates. Rising power, chip and data-center costs are adding pressure just as frontier models demand ever larger computational budgets.