Anthropic‘s Opus 5 model generated just 3.5 per cent of the company’s total billing volume in July 2026, trailing significantly behind older versions despite being the most capable system. The launch occurred on 24 July, yet spending on the flagship model remained low while cheaper tools like Sonnet 4.6 and Opus 4.8 accounted for nearly half of the total usage. This data comes from the Ramp AI index, which tracks spending across 70,000 business credit cards. The trend suggests that enterprise clients prioritise cost efficiency over raw capability when selecting large language models for daily workflows. OpenAI reported a 35 per cent revenue increase in the same period, driven partly by the performance of its GPT 5.6 release. While Anthropic expects to become profitable in the third quarter with 6,000 high-value customers, the disparity between model quality and actual adoption highlights a clear market preference for value. Companies are willing to accept slightly lower performance if it reduces operational expenses. The shift indicates that the era of paying a premium for the absolute best model is ending in favour of sustainable, budget-conscious deployment strategies.
* Opus 4.8 led spending at 28.0 per cent
* Fable 5 received 8.0 per cent of total spend
* Top five models combined for 53.4 per cent of usage
