Financial technology unicorn Revolut turns to Zhipu GLM, programming bills reduced to about one-tenth

The European highest-valued fintech unicorn Revolut has set the GLM of the AI company Zhipu as the default AI programming model for approximately 2,000 engineers, with tools like Claude Code and Codex relegated to alternatives, mainly reserved for complex tasks. David Abitbol, the head of AI operations at Revolut, revealed that the company previously set a limit of $300 per person per month for Claude Code, but it provided limited help to overall efficiency.

In April this year, the company began to deploy GLM independently and opened unlimited Tokens to engineers. Currently, Revolut uses about 56 NVIDIA B200s to run GLM 5.3, with a comprehensive cost of approximately $0.30 per million Tokens, and the related bill is only one-tenth of what was previously spent on Anthropic and OpenAI.

According to internal company statistics, the coding time metrics for engineers improved by about 60%, and the cycle metrics from task handover to delivery improved by about 70%. In the past, with limited Tokens, engineers tended to reserve their quotas for complex tasks; now that there are no usage restrictions, they have begun to extensively use AI to search internal documents, handle Jira tickets, and find requirement specifications in Slack.