Terminal X's Eval Agent on GPT-6.1 Sol agrees with senior investors 88.6% of the time

Turning investor judgment into agents a firm owns

Terminal X targets what it calls memory debt: an investment firm may keep its documents but lose the reasoning behind decisions when senior investors leave. Its Capital Memory operating system is designed to pay that debt down.

The company's forward-deployed engineers, described as former finance and investment professionals at JPMorgan and Blackstone, work with each customer's investment teams. They design the firm's agentic architecture and encode how the firm invests into agents that the firm owns, deployed on-premise or in its own cloud. OpenAI models, including GPT-6.1 Sol and GPT-6 Luna, handle the reasoning inside those agents.

Engineers spend weeks onsite reverse-engineering trusted memos and capturing processes that were never written down, such as how a credit desk screens a borrower. These are encoded in hundreds of agents, customized by asset class and strategy, with sub-agents for each investment stage.

How the models split the work

GPT-6.1 Sol handles the stages that need judgment and a very large context window: synthesizing the investment thesis, identifying key considerations and assessing fit with the firm's mandate. GPT-6 Luna runs the high-volume steps where token efficiency matters more than depth.

The agents operate inside each firm's private data room and process documents in English, Korean, Japanese and Chinese. What they learn is stored as a firm Memory, a versioned record of how one desk makes one kind of decision, including past analyses, committee decisions, workflow instructions and house views. Every production query is auditable, with the steps taken, sources retrieved, model used and reasoning.

According to Terminal X, one private equity firm reported $2.4 million in annualized savings from automated due diligence and investment memo workflows.

The Eval Agent and its 54-criterion rubric

To check output against a firm's own standard, Terminal X built an Eval Agent, an orchestrated workflow on GPT-6.1 Sol and other OpenAI models. It is designed to read agent output the way a VP reads an analyst's work, scoring each result against a 54-criterion rubric. Checks ask whether numbers, comps, consensus figures and quotes come from the right sources, whether the narrative holds, and whether the recommendation ties to the house view.

Each check scores pass or fail. Every fail includes a one-line reason and a better source or corrected text, and becomes a Memory, for example a new workflow instruction or test case for that desk. Ken Lee, Co-Founder and CTO, says: "Every answer gets eval'd, and whatever the analyst corrects teaches the system a little more about how that firm works."

Benchmark results: style drift and cost

Terminal X says it picks models based on its own evaluation results. Across 500 scored judgments, the Eval Agent on GPT-6.1 Sol agreed with the senior human reviewer 88.6% of the time, versus 86.4% for the next-best model, an improvement of 220 basis points.

One error class where the model did better is style drift. In the source's example, a buyout fund's mandate is to find take-private candidates trading at a discount to NAV, but when scenario models suggest a market turn, deals start recommending a distressed credit position or a rescue financing. The Eval Agent on GPT-6.1 Sol caught about 25% more of these departures than the next-best model.

Token efficiency on long-context work let Terminal X build a deeper review within the same budget. The thesis is tested against the mandate, then against recent deals, and uncertain verdicts are re-run and reconciled. According to the source, this flagged deals that a single pass on a comparable model had waved through. The same workflows ran at about 30% lower cost per query than the next-best model, at similar speed.

What it means for investment firms

Using GPT-6.1 Sol and GPT-6 Luna, Terminal X reports 1,700 firm Memories, more than 1 million custom workflows, and a weekly active user count equal to 81% of its monthly active users. Its next step is to extend Memories from desks to individuals, so each professional's standard is captured as it is practiced.

For firms, the pitch is that judgment becomes measurable and auditable data they own. CEO Hyun Hong put it this way: "The firms that win the next decade will own their judgment as data." The performance figures come from Terminal X's own evaluations as reported by OpenAI, not from independent testing.