Revolut has made Zhipu’s GLM its default AI coding model for about 2,000 engineers, with related costs falling to roughly one-tenth of its previous bill. According to ChainCatcher, Claude Code and Codex are now mainly reserved for complex tasks at the European fintech unicorn, which has the continent’s highest valuation.
Revolut AI operations lead David Abitbol said the company previously set a $300 monthly Claude Code allowance per engineer, but the tool had a limited effect on overall efficiency. In April this year, Revolut began running GLM in-house and gave engineers unlimited tokens. The company now runs GLM 5.3 on about 56 Nvidia B200 GPUs, at an estimated combined cost of $0.30 per million tokens, compared with its previous spending on Anthropic and OpenAI tools.
Internal company metrics show coding-time indicators improved by about 60%, while the time from taking on a task to delivering it improved by about 70%. With token use no longer capped, engineers have also begun using AI more extensively to search internal documents, handle Jira tickets, and find requirements in Slack.