Revolut has replaced Claude with self-hosted models, reducing inference costs by up to 90% in less than a year.According to infrastructure lead David Abitbol, the setup runs on 56 Nvidia B200 GPUs and brought inference down to about $0.30 per million tokens.The shift also led to a 70% improvement in lead time productivity and a 60% jump in coding time productivity, though self-hosting requires heavy hardware…https://en.hacks.gr/i-revolut-ekopse-kata-90-to-kostos-tis-afinontas-ta-ergaleia-technitis-noimosynis-tis-claude/#Revolut #ArtificialIntelligence #Nvidia #MachineLearning
Revolut has replaced Claude with self-hosted models, reducing inference costs by up to 90% in less than a year.According...
