💸Revolut Cuts AI Coding Inference Costs 90% With Self-Hosted GLM Models

Your AI coding bill could drop 90% this way

Revolut just ran AI coding at scale on its own models, and the bill fell roughly 90%. The fintech made self-hosted models its default. Yes, really. If you're paying per token for Claude Code or Codex, this should make you itchy. Revolut's team found that self-hosting GLM 5.1 cost about a tenth of the big API options. Debugging and wiring the model into Jira and Confluence tickets consumed the effort. Code generation was the easy part, and most teams underestimate that gap. The numbers: Revolut rolled AI coding tools out to roughly 2,000 engineers and saw 60% faster coding time and 70% faster lead time. It now serves GLM-5.3 on 56 Nvidia B200 GPUs, with sub-3-second time to first token at around 256K context. Engineers get about 200 tokens per second, and around 800 after 5 pm. The catch? GLM is text only, so UI work still goes to Gemini or GPT through a model router.