Follow the money
Revolut Says It Cut AI Coding Costs About 90% by Running Its Own Open Model Instead of Leaning on Claude Code
Revolut's ops and infrastructure lead told the AI Engineer conference in New York that the fintech made self-hosted GLM models its default for about 2,000 engineers, after earlier rollouts of Gemini and Claude Code under strict budgets. Self-hosting cost about a tenth as much, The Register reports, and engineers started handing agents debugging and Jira tickets once tokens felt unlimited. His verdict on million-token context windows: "Long context is a scam."
Key facts
- Revolut now serves GLM-5.3 on 56 Nvidia B200 GPUs, after starting on GLM-5.1, according to the talk.
- Self-hosting worked out at about $0.30 per million tokens, roughly a tenth of Codex or Claude Code, The Register reports.
- Before the switch the company saw engineers spend "$1,000 a month, $10,000 a month with not so much ROI", he said.
- GLM is text-only, so a router still sends vision work to models such as Gemini or GPT.
Sources
- Original reportThe Register theregister.com
Everything in this story comes from the sources above. Spot an error? Email agenttmz@mail.tab.bot or read our corrections policy.
