Revolut PRAGMA

Revolut's in-production transaction foundation model, serving 80 million customers on NVIDIA Blackwell Ultra GPUs.

Revolut · Enterprise and vertical AI applications · model

Published in Feed

Overview

PRAGMA is a single transaction foundation model that Revolut is using to rebuild its technology stack, described in an NVIDIA GTC Berlin post published October 21, 2026. It is built with NVIDIA cuDF and Dynamo-Triton on Blackwell Ultra GPUs, is in production serving an initial 80 million customers, and is engineered to scale 3x. The post gives no parameter count, pricing or general availability date and does not describe external access, so it is catalogued as a bank-internal system.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • Single foundation model trained on transaction data
  • In production for an initial 80 million customers
  • Engineered to scale 3x
  • Built with NVIDIA cuDF and Dynamo-Triton on Blackwell Ultra GPUs

Use cases

  • Powering Revolut's rebuilt technology stack for payments and banking features
  • Reference for other banks building transaction models (Lloyds Banking Group's ATUM is a comparable model in the same post)

Pricing

Not applicable or unknown; no external pricing or access is described.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
Not listed
Record updated
27 Oct 2026
Record last checked
27 Oct 2026

Explore the Feed

Other entries filed under enterprise and vertical ai applications; inclusion does not imply equivalent features.

Browse the Feed