The benchmark addresses a persistent gap in AI evaluation by moving beyond simple data extraction. FrontierFinance includes 220 diverse queries and over 11,000 expert-crafted rubrics that span the entire investment lifecycle, from initial idea screening and macro analysis to company research and catalyst monitoring. According to the company, the benchmark is built on a foundation of 5,000 complex examples developed over years of designing AI agents for high-stakes financial environments.

In head-to-head testing, Samaya’s high-effort configuration achieved 56% accuracy, outperforming Claude Fable 5, which recorded 49.2%, and GPT-5.6 Sol at 46.8%. Beyond raw accuracy, the company emphasizes the efficiency of its system, reporting that its low-effort mode reached 50.8% accuracy at four times the cost-efficiency of the leading Anthropic model. Maithra Raghu, CEO and Founder of Samaya AI, noted that the platform was designed to address the unique intolerance for error in finance, where near-correct responses often result in failed investment outcomes.