The Register reports that Revolut says it cut its AI coding inference costs by roughly 90 percent in under a year by making self-hosted models its default rather than relying mainly on commercial services such as Claude. Speaking at the AI Engineer conference in New York, ops and infrastructure lead David Abitbol said the fintech rolled AI coding tools out to about 2,000 engineers, adopted Z.ai's open-weights GLM-5.1 and now serves GLM-5.3 on 56 Nvidia B200 GPUs. Self-hosting cost about $0.30 per million tokens, roughly a tenth of Codex or Claude Code, and he said cheap tokens let engineers hand the AI debugging and ticket work, which is where the biggest gains came from. He also called very long context windows "a scam" in practice.

About 1,900 upvotes and 340 comments in a 329,000-member community. The top reply says this is why running models locally is not a waste of time; another argues that self-hosting means a company knows exactly which model it is running, while a cloud provider can change it without notice. One commenter who says they know people at Revolut describes its engineering culture as cautious.