Revolut moves about 2,000 engineers onto self-hosted GLM for coding, at about a tenth of its Anthropic and OpenAI bill

Revolut's head of AI operations David Abitbol told the AI Engineer conference that the neobank now runs Z.ai's open-weight GLM model on its own infrastructure as the default coding model for about 2,000 engineers, keeping Claude Code and Codex as exceptions. About 56 Nvidia B200 GPUs serve the model at a blended 30 cents per million tokens, an invoice roughly a tenth of what it paid Anthropic and OpenAI. Since self-hosting began in April, coding time improved about 60% and lead time about 70% on its own figures, which Abitbol links to engineers no longer rationing tokens. The weights run on US-based infrastructure rather than through the developer's endpoint in China.