Samaya’s finance agent beats analyst baselines, challenging equity-research workflows
A controlled 456-company test puts live financial data—not model memory—at the center of the performance gain.
In a vendor-run test covering 456 Q2 earnings releases, seven frontier models used Samaya’s point-in-time financial-research harness to forecast revenue, gross margin, operating income and adjusted earnings per share five trading sessions before results. Samaya reports that the latest models beat both raw analyst consensus and its stronger bias-corrected baseline on several measures, and says an alpha version is available inside its product for users and clients. If the task-specific results transfer to live investment work, specialist information providers may capture more value through proprietary data-and-agent systems while routine forecasting loses some of its scarcity.