Santander’s October 19 announcement sets out a three-year collaboration framework with NVIDIA to develop and evaluate AI models and agents for banking. The added technical detail includes NVIDIA Kumo for specialized intelligence in core-banking processes, NVIDIA Dynamo-Triton for integrating capabilities into the bank’s infrastructure, and controlled testing of conversational AI before potential deployment across the group. Santander’s announcement describes development and evaluation work, not a completed bank-wide rollout.
What the named technologies do
Santander’s Model Arena evaluates open-weight models against banking tasks, helping the bank assess which model fits a particular use. The bank is also developing the Arena to design, simulate, evaluate and monitor banking agents. NVIDIA NeMo Customizer is used to adapt models for specific tasks.
Santander says it has developed specialized intelligence for core-banking processes using NVIDIA Kumo, a family of open-weight foundation models for structured data. It says NVIDIA Dynamo-Triton is being used to integrate these capabilities into its technology infrastructure and enabled the cost reduction described in the announcement. Each tool has a different role: model evaluation, task-specific adaptation, structured-data modeling and infrastructure integration.
How Santander plans to test banking agents
The Model Arena is also intended to help Santander test agents before possible deployment. Santander says conversational AI is stress-tested in controlled environments and that governance mechanisms are applied to agents to improve control and security. That testing process is part of the announced development program; the collaboration does not establish that these systems are already operating across the bank.
The framework also names foundation models, agents, fraud and anti-money-laundering capabilities, real-time decision-making and banking digital twins as collaboration areas. These describe the work Santander and NVIDIA plan to pursue, rather than a list of completed services. For readers following the financial side of technology, NTV88’s finance coverage offers a broader companion read.
What Santander’s experimental cost figures mean
Santander says selected experiments lowered execution costs from about €5 to about €0.03 per interaction while maintaining good results. Calculated from those approximate figures, the implied reduction is about 99.4%. Santander separately reports improved conversation quality and a cost reduction of more than 50% in one experimental customer-service use case. That is a distinct result, not the same comparison as the per-interaction figures.
