British fintech Revolut cut AI coding inference costs by roughly 90% in less than a year by moving away from commercial assistants toward self-hosted open-weight models. Deployed to about 2,000 engineers, the tools reduced coding time by 60% and lead time from task initiation to delivery by 70%. Revolut now runs Z.ai’s GLM-5.1 and GLM-5.3 models on 56 Nvidia B200 GPUs, at a blended cost of approximately $0.30 per million tokens. Optimizations carried out from April to August 2026 pushed throughput above seven requests per second and reduced time to first token to 1.6 to 1.68 seconds. The models run on infrastructure controlled by Revolut, while code suggestions remain subject to human review to maintain regulatory compliance.
Source: Read the original article
