Revolut says self-hosted GLM gave its engineers unlimited AI coding at about a tenth of the cost

In two October conference talks, Revolut's AI infrastructure lead David Abitbol said the bank has served the open-weight GLM model from Z.ai to about 2,000 engineers since April. It runs on roughly 56 Nvidia B200 GPUs, at a blended 30 cents per million tokens. He said that is about a tenth of what Revolut spent on Anthropic and OpenAI, with 70% better lead time. Frontier models such as Claude and Codex are now kept for the hardest problems.

Why it mattersIt's a rare, detailed account of a big regulated company swapping frontier APIs for a self-hosted open model. But every number is Revolut's own and unaudited, the "tenth" is described differently in the two talks, and the 70% gain arrived when Revolut lifted engineers' token budgets, so it isn't proof GLM beats Claude. Self-hosting took months of tuning. Our check rates the viral post partly fact. We haven't tested GLM hands-on.

Curated by Sara Simeone as part of our human editorial coverage. The original sources are linked below; curation is distinct from a Hype or Fact? assessment.

SHARE THISPost on X ↗Share on LinkedIn ↗Copy link

How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.