Revolut says it cut AI coding inference costs by roughly 90 percent in less than a year. The move: drop commercial tools like Claude Code as the default and replace them with self-hosted GLM models.
What Changed
The British fintech shifted its default AI coding setup away from commercial API-based tools toward self-hosted GLM models running on its own infrastructure. According to the company, the inference cost reduction hit approximately 90 percent over the course of less than twelve months.
The Operator Takeaway
For any company running AI coding assistants at scale, inference costs compound fast. Revolut’s move signals that self-hosted open models are now capable enough to displace commercial offerings as the daily driver for engineering teams, not just as a cost-cutting experiment. The tradeoff is infrastructure ownership and the engineering overhead that comes with it. For a fintech at Revolut’s size, that math apparently works out.
Smaller teams should watch this closely. The same cost-per-token gap that made self-hosting worthwhile for Revolut will show up at smaller scale too, just with a higher relative setup cost to absorb first.
