How Leaders in the Financial Sector Cut AI Coding Costs 90% by Self-Hosting
Revolut cut AI coding inference costs by approximately 90% in less than a year, replacing commercial AI coding services with a self-hosted deployment of Z.ai’s GLM-5.3.
The deployment serves roughly 2,000 engineers through 56 Nvidia B200 GPUs at $0.30 per million tokens, compared with monthly commercial seats that previously cost $1,000 to $10,000 per engineer. Revolut reports a 60% improvement in coding time and a 70% improvement in lead time, with sub-three-second time to first token.
- ~90%
- cut in AI coding inference costs, in less than a year
- $0.30
- per million tokens, compared with monthly commercial seats that previously cost $1,000 to $10,000 per engineer
- 60%
- improvement in coding time, as reported by Revolut
- 70%
- improvement in lead time, as reported by Revolut
The problem
David Abitbol, Revolut’s operations and infrastructure lead, said: “We were seeing engineers spend $1,000 a month, $10,000 a month with not so much ROI.” Engineers had been spending thousands monthly on Claude Code and OpenAI’s Codex, and Emergent describes the monthly bills as unpredictable.
What they did
Revolut serves roughly 2,000 engineers through 56 Nvidia B200 GPUs, running a 754-billion-parameter mixture-of-experts model with 40 billion parameters activated per token.
The result
Time to first token is under three seconds, with throughput near 200 tokens per second during business hours. David Abitbol, Revolut’s operations and infrastructure lead, said: “The billing is flat, we don’t have any bill shock at the end of the month.” Revolut caps recommended context length at 256,000 tokens.
Figures come from the source below. Anything from another place is attributed in the text.
Source: Emergent, reporting The Register: Revolut cuts AI coding costs 90% with self-hosted GLM-5.3 ↗