Terminal X achieved 88.6% alignment with investors using GPT-6.1 Sol-based Eval Agent

Terminal X announced that the GPT-6.1 Sol-based Eval Agent reached the same conclusion as senior human evaluators at a rate of 88.6%. The company aims to reduce reasoning loss in investment decisions with the Capital Memory operating system.

To address the issue of decision reasoning loss faced by investment firms, referred to as memory debt, Terminal X developed the Capital Memory operating system. Field engineers experienced with JPMorgan and Blackstone design agent architectures in collaboration with clients' investment teams. During this process, OpenAI models GPT-6.1 Sol and GPT-6 Luna handle the reasoning function of the agents, while Terminal X's Eval Agent scores the outputs against standards. Thanks to GPT-6.1 Sol, the Eval Agent catches proposals deviating from the fund framework approximately 25% more often, while delivering about 30% lower cost per query compared to the closest competing model.

Based on 500 scored evaluations, the Eval Agent using GPT-6.1 Sol reached the same conclusion as senior human evaluators 88.6% of the time, showing a 220 basis point improvement compared to the closest competing model's rate of 86.4%. While Hyun Hong serves as CEO and Ken Lee as CTO, the system has created 1,700 corporate Memory records to date, executed over 1 million private workflows, and brought the weekly active user rate to 81% of monthly active users.