Revolut shifts 2,000 engineers to self-hosted GLM for coding

Revolut moved roughly 2,000 engineers to a self-hosted GLM model for coding tasks, dropping frontier models like Claude Code and Codex to an exception list. According to AI ops lead David Abitbol, the migration utilized 56 Nvidia B200s and reduced token costs to 30 cents per million tokens, resulting in a 70% improvement in engineering lead time as engineers stopped rationing tokens. Open article →