SigmaCore Study Finds Standard AI Gets Finance Math Wrong 36% of the Time
A new benchmark from SigmaCore.ai exposes a 63.5% accuracy rate for standard AI in financial math, as the firm touts its own deterministic engine.
Standard generative AI models answer financial mathematics questions correctly only 63.5% of the time, according to benchmark research released by Alpharetta, Georgia-based SigmaCore.ai, an enterprise developer of deterministic AI systems.
The study highlights what SigmaCore characterizes as structural calculation flaws inherent in conventional large language models when applied to quantitative financial tasks — a gap the company says poses meaningful risk for institutions relying on AI-assisted analysis.
Read more IGT and MGM Springfield Debut Massachusetts' First Wide-Area Progressive Slots →
SigmaCore contends its own deterministic calculation engine achieves 100% precision on the same battery of tests, positioning the product as a purpose-built alternative to general-purpose generative models in finance-specific workflows.
The findings arrive as financial services firms accelerate adoption of AI tools for tasks ranging from portfolio analysis to regulatory reporting, raising industry-wide questions about the fitness of general AI architectures for high-stakes numerical work where rounding errors or probabilistic outputs carry real consequences.
Analysts and independent researchers have not yet publicly verified SigmaCore's methodology, and the benchmark was self-commissioned — a standard caveat for vendor-released studies. Continue reading at All Financial Services & Investing
