Revolut says self-hosting GLM cut its AI coding costs by about 90%

Revolut says it cut its AI coding inference costs by roughly 90 percent in under a year by making self-hosted open models its default instead of relying mainly on services like Claude, The Register reports from a talk by ops and infrastructure lead David Abitbol at the AI Engineer conference in New York. Its roughly 2,000 engineers now use Z.ai's GLM-5.3, served on 56 Nvidia B200 GPUs, with a router sending vision tasks to models like Gemini or GPT Luna because GLM is text only. Abitbol said the biggest gains came from cheap tokens letting engineers hand debugging and Jira and Confluence tickets to AI, and that about 256,000 tokens of context is what they really need, not one million. The savings figure is Revolut's own and was not independently checked.