Terminal X says its AI investment agents agree with senior reviewers 88.6% of the time

Terminal X says its investment agents agreed with senior human reviewers on 88.6% of 500 scored judgments, compared with 86.4% for the next-best model in its test. The startup, which builds AI systems for investment firms, says its evaluation agent also caught about 25% more recommendations that strayed from a fund’s mandate and ran the same workflows at roughly 30% lower cost per query.

The figures were published by OpenAI in a customer profile dated October 20, 2026, and reflect Terminal X’s own evaluation rather than an independent study. OpenAI says the system uses GPT‑6.1 Sol and GPT‑6 Luna, with the former handling tasks requiring deeper reasoning and the latter used for higher-volume work where token efficiency matters more.

Terminal X’s Capital Memory product is designed to turn a firm’s past decisions and unwritten processes into reusable workflows. The company says its engineers work with investment teams to review trusted memos and practices, then build agents tailored to asset classes and stages of an investment process. Those agents operate in customers’ private data environments, and the resulting workflows can retain a record of sources, steps, models and reasoning for each query.

A key part of the system is an evaluation agent that scores outputs against a 54-criterion rubric, checking matters such as whether figures and quotes come from appropriate sources and whether a recommendation fits the firm’s stated investment approach. Terminal X says failed checks include a reason and a correction or better source, which can then be used to update the firm’s stored instructions or tests.

The company says one private-equity customer reported $2.4 million in annualized savings from automating due diligence and investment-memo workflows. Terminal X also reports that it has built 1,700 firm “Memories,” run more than 1 million custom workflows and reached a weekly active user count equal to 81% of its monthly active users. These are company-reported figures; the OpenAI profile does not provide independent verification or detail on how the savings were calculated.

Terminal X says it plans to extend its Memories from investment desks to individual professionals, capturing how each person works in addition to the practices of a team. The broader pitch is that firms can preserve and reuse investment judgment as staff and teams change, though the reported agreement rate and mandate-detection results do not by themselves establish how the agents perform across firms or in live investment outcomes.