ProCap Financial, Inc. (NASDAQ:BRR) (the "Company"), the first publicly traded agentic finance firm, today announced that Silvia, its AI agent lab exclusively focused on finance, has achieved industry-leading performance, outperforming frontier AI models evaluated on important personal finance topics including tax, mortgage, and credit cards.
Over the past six months, Silvia’s engineering team rebuilt its technology stack to own its own intelligence rather than rely on off-the-shelf, generic AI tools. The result is a custom system optimized exclusively for personal finance use cases, delivering higher accuracy, primary-source citations, faster response times, and lower costs.
Key results include:
- Silvia ranked #1 among seven AI products on the Company’s public tax, mortgage, and credit card benchmarks for both accuracy and citing primary sources. The benchmarks are open sourced on Hugging Face.
- Since the new model began rolling out in July, median response time has dropped 48%, cost per question has dropped 59%, and the share of answers users rate negatively has fallen by more than half, while monthly active users have grown substantially since March.
The proprietary technology stack consists of three core components:
- Data pipelines: Custom benchmarks graded on accuracy and primary-source citations, plus high-quality training data generation.
- Harness engineering: Tight integration of the model router, tools, and agents so the system operates as a unified whole.
- Model training: Silvia Analyst, a quantitative finance specialist model that handles market data, company financials, portfolio analysis, and document parsing at 97% lower cost than frontier models.
Login to comment