Revolut cuts AI coding costs by 90% with self-hosted models
Summarised from 2 outlets · Updated 4 Nov, 10:32 · Archive
British fintech Revolut has reduced its artificial intelligence infrastructure costs by around 90% by replacing paid services with self-hosted open-weight models.
Revolut has reduced its reliance on commercial AI services by switching to self-hosted, open-weight models from China's Z.ai. The company previously used Anthropic's Claude but found that heavy users were spending thousands of dollars per month on tokens without generating equivalent return on investment.
Revolut Infrastructure Lead David Abitbol said the company was 'highly skeptical about the cost of AI' and had launched with strict budget limits. The switch to self-hosted models has brought inference costs down to around $0.30 per million input and output tokens, roughly one-tenth the cost of similar commercial services.
The change has improved developer productivity. Abitbol noted that the self-hosted option resulted in a 70% improvement in lead time productivity and a 60% improvement in coding time productivity. With cheaper tokens available, engineers now tackle more tasks including debugging and clearing Jira tickets.
TechRadar reports that while direct model usage costs have been slashed dramatically, self-hosting still requires significant outlay in hardware and infrastructure to manage the company's own inference capacity.

How it is being reported
- 'We were highly skeptical about the cost of AI': Revolut reveals it cut AI coding costs by cutting out Claude entirelyRevolut ditches Claude as it lays out a plan to use self-hosted models – but infrastructure costs are still eye-watering.TechRadar · 4 Nov, 10:10
- British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hostedTurns out cheap tokens get engineers doing things that save time, like debugging and tackling Jira ticketsThe Register · 3 Nov, 20:07
In this story: Revolut · Anthropic · Z.ai · Google · David Abitbol · London · China
This summary was written by AI from the headlines and standfirsts above, and states as fact only what at least two outlets report. How we use AI · Report a problem · Corrections log
Comments (0)