Terminal X audits investment teams’ decision memory with GPT‑6.1
Terminal X’s Capital Memory system turns investment-team processes into private agents, while an Eval Agent scores outputs against 54 criteria.
AT A GLANCE
Key points
- The Eval Agent scores every output against 54 criteria
- GPT‑6.1 Sol matched human reviewers 88.6% across 500 judgments
- Terminal X reports 25% higher detection of off-mandate recommendations
What is changing?
OpenAI detailed Terminal X’s Capital Memory system on October 20. Terminal X engineers encode investment teams’ unwritten review practices into agents that run inside each firm’s own data room. GPT‑6.1 Sol handles long-context stages such as investment-thesis synthesis and mandate fit, while GPT‑6 Luna is used for higher-volume steps.
The company’s Eval Agent checks every output against 54 criteria covering sources, figures, comparables, narrative quality, and the firm’s house view. According to Terminal X, GPT‑6.1 Sol agreed with senior human reviewers 88.6% of the time across 500 scored judgments, caught roughly 25% more recommendations that strayed from a fund’s mandate than the next-best model, and reduced per-query cost by about 30%.
Terminal X stores the model used, retrieved sources, and workflow steps in an audit trail. These figures come from Terminal X’s customer work and should not be interpreted as an independent measure of general investment performance.
Explore the original sources
Source date: 20 October 2026.
Published by AI Radar: 21 Oct 2026, 05:00 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…