Revolut built an AI model called PRAGMA (PRe-trained Banking Foundation Model), a series of transformer-based models developed with Nvidia for fraud detection, credit scoring, and other financial tasks. PRAGMA was trained on roughly 40 billion banking events from approximately 25 million Revolut users across 111 countries, amounting to 207 billion tokens. On internal benchmarks, the model showed a 64.7% improvement in fraud recall and a 16.7% increase in fraud precision compared to existing systems, while credit scoring achieved up to a 130% uplift in PR-AUC metrics. PRAGMA runs on Nvidia’s H100 GPUs within the Nebius AI Cloud infrastructure, with model sizes ranging from 10 million to 1 billion parameters; the largest variant, PRAGMA-L at 1 billion parameters, used up to 64 Nvidia H100 GPUs. The research paper was submitted on April 9, 2026, and presented at Nvidia’s GTC 2026 conference. Revolut reports the approach decreased its dependency on traditionally engineered features.