Revolut moves internal coding agents to open-weight GLM
The $115 billion fintech runs GLM across 56 Nvidia B200 chips, reporting lower token expenses and faster delivery cycles.
Fintech firm Revolut has transitioned its developer coding assistant setup to the GLM open-weight model, shifting workload execution to an in-house cluster [1]. The company, recently valued at around $115 billion, employs roughly 2,000 software engineers [1].
To operate the open model internally, Revolut deployed 56 Nvidia B200 graphics processing units [1]. According to reporting on the company's internal figures, running the workload directly costs 30 cents per million tokens [1]. Following the transition, Revolut observed a 60% decline in hands-on coding duration and a 70% contraction in overall engineering lead time [1].
David Abitbol, head of AI operations and infrastructure at Revolut, pointed out that software bottlenecks extend beyond raw syntax generation, noting that "the bottleneck for engineering is not necessarily coding" [1]. The transition reflects a corporate shift toward self-hosted open-weight architectures to reduce reliance on commercial coding interfaces and contain infrastructure costs.
Questions and answers
What hardware supports Revolut's GLM deployment?
The fintech company runs the GLM model using 56 Nvidia B200 accelerators.
What token costs did Revolut report for running GLM?
Revolut registered an operating expense of 30 cents per million tokens.
How did developer efficiency change after the switch?
Engineering lead times dropped by 70%, while hands-on coding time dropped by 60% across roughly 2,000 developers.
Companies & investors in this story
Sources
- 1.Revolut Cuts A.I. Coding Costs 90% With Open-Weight GLM | Trending Topics — trendingtopics.eu
- 2.Revolut makes self-hosted GLM its default AI coding model — Digg, Nov 3, 2026
- 3.Revolut Says Self-Hosted Open AI Models Cut Its Coding Costs 90% — Startup Fortune, Nov 3, 2026
