Revolut cut its costs by 90% by leaving Claude's AI tools
Revolut cut Claude for self-hosted models, slashing inference costs by up to 90% in under a year. Running on 56 Nvidia B200 GPUs, the setup reduced inference to around $0.30 per million tokens while improving coding pro
Revolut cut Claude for self-hosted models, slashing inference costs by up to 90% in under a year.
Running on 56 Nvidia B200 GPUs, the setup reduced inference to around $0.30 per million tokens while improving coding productivity by 60%.
How the situation has changed for developers
Before the change, Revolutβs most active users were spending thousands of dollars each month on digital access keys for its AI tools. The companyβs head of infrastructure, David Abitbol, ββsaid the option put an end to restrictions and high bills.
With the new systems, the cost has dropped to about $0,30 per million input and output data. Now, developers can work freely, without worrying about unexpected costs.
Equipment and productivity benefits
To run the new system locally, Revolut is leveraging its own hardware, specifically 56 dedicated Nvidia B200 processors with over 380GB of memory. This move, however, requires additional investments in infrastructure, maintenance, and electricity.
β³
READ ALSO
HACKS.GR / RELATED
AI Unlocked
Waymo to end partnership with Uber in Austin and Atlanta by 2028
β
This change is also paying off in the daily work of developers. According to the company's data, the speed of task completion has improved by 70%, while the time spent writing code has increased by 60%.
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
β
end of article
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.