Terminal X Capital Memory on GPT-6.1 Sol: 88.6% Agreement With Senior Investor Judgment

Terminal X's Capital Memory system, built on OpenAI's GPT-6.1 Sol and GPT-6 Luna, matched senior human investor judgment 88.6% of the time across 500 scored judgments, compared with 86.4% for the next-best model — a 220 basis point improvement. The result matters because it is a concrete measurement of whether an agentic system can encode a specific investment firm's judgment, not just retrieve its documents, and it is paired with a reported ~30% lower cost per query on the same workflows.

What Terminal X announced and why it matters

Terminal X operates a product called Capital Memory, described as an operating system for preserving and applying an investment firm's decision-making. The company's framing is that when senior investors leave a firm, the firm keeps the documents but loses the reasoning — a problem Terminal X calls memory debt. Capital Memory is designed to pay that debt down by encoding how a firm invests into agents the firm owns, deployed on-premise or in the firm's own cloud.

OpenAI models handle the reasoning inside those agents. Terminal X's Eval Agent grades the agents' output against the firm's standards. According to the announcement, GPT-6.1 Sol helped the Eval Agent catch about 25% more recommendations that strayed from a fund's mandate, and ran the same workflows at approximately 30% lower cost per query than the next-best model, at similar speed.

How Terminal X builds agents that invest like a specific firm

Terminal X builds a custom agent architecture for each customer rather than shipping a single general-purpose investing agent. Its forward-deployed engineers — described as former finance and investment professionals from JPMorgan and Blackstone — spend weeks onsite with the customer's investment teams, reverse-engineering the firm's trusted memos and capturing processes that were never written down, such as how the credit desk screens a borrower or which sources the co-investment team pulls before a memo.

Those processes are encoded in hundreds of agents, customized by asset class and strategy, with sub-agents for each investment stage. The division of labor between models is explicit:

GPT-6.1 Sol handles the stages that require judgment and a 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 work inside each firm's private data room, a dedicated environment for enterprises to share and manage sensitive investment data. They process documents in English, Korean, Japanese, and Chinese, and store what they learn 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. Engineers then turn each Memory into a repeatable workflow on Terminal X's platform, where every production query is auditable with its full trail — the steps taken, sources retrieved, model used, and the reasoning.

Terminal X reports that customers see significant savings from this approach, including one private equity firm that reported $2.4 million in annualized savings from automated due diligence and investment memo workflows.

The Eval Agent: a 54-criterion rubric applied to every output

Terminal X built an Eval Agent to check agents' answers against the firm's own standard. It is an orchestrated workflow on GPT-6.1 Sol and other OpenAI models, designed to read every agent output the way a vice president reads an analyst's work.

The Eval Agent checks each output against a 54-criterion rubric, asking whether numbers, comps, consensus figures, and quotes were pulled 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 plus a better source or corrected text, and becomes a Memory. A missed source, a wrong comparable, or a thesis that drifts from the house view is encoded as a new workflow instruction or a new test case for that desk.

"Every answer gets eval'd, and whatever the analyst corrects teaches the system a little more about how that firm works. So the longer a firm uses it, the more it thinks like their own team." —Ken Lee, Co-Founder and CTO of Terminal X

Benchmark results: 88.6% agreement and fewer mandate departures

Terminal X selects models based on its own evaluation results rather than vendor benchmarks. Across 500 scored judgments, its 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 GPT-6.1 Sol outperformed is style drift. The example given: a buyout fund's mandate is to find take-private candidates trading at a discount to NAV, underwrite the business, and structure the financing. When scenario models suggest a market turn, deals start recommending a distressed credit position or a rescue financing instead. The Eval Agent on GPT-6.1 Sol caught about 25% more of these departures than the next-best model did.

GPT-6.1 Sol's token efficiency on long-context work let Terminal X build a deeper review workflow within the same budget. In that workflow, the thesis is tested first against the fund's mandate, then against the fund's recent deals; uncertain verdicts are re-run, and the agent reconciles the results before reporting. Terminal X states that this deeper review flagged deals that a single pass on a comparable model had waved through.

"GPT-6.1 Sol's long-context reasoning and token efficiency allow us to build deeper evaluation pipelines. This helps our Eval Agents make investment judgment measurable and auditable." —Hyun Hong, Co-founder and CEO, Terminal X

Scale and next steps

Using GPT-6.1 Sol and GPT-6 Luna, Terminal X reports 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.

The company's stated next step is to extend firm Memories from desks to individuals, so that each investment professional's standard is captured as it is practiced, not only as it is described.

"The firms that win the next decade will own their judgment as data. Our job is to make that judgment measurable, so it can be kept, improved, and reused." —Hyun Hong, Co-founder and CEO, Terminal X

Key facts at a glance

Models used: GPT-6.1 Sol (judgment-heavy, long-context stages) and GPT-6 Luna (high-volume, token-efficiency-sensitive steps).
Eval Agent agreement: 88.6% with the senior human reviewer across 500 scored judgments, versus 86.4% for the next-best model (220 basis point improvement).
Mandate-drift detection: about 25% more style-drift departures caught than the next-best model.
Cost: approximately 30% lower cost per query than the next-best model, at similar speed.
Evaluation rubric: 54 criteria per output, with pass/fail scoring and a one-line reason plus corrected source for each failure.
Deployment: on-premise or in the customer's own cloud, inside a private data room.
Languages processed: English, Korean, Japanese, and Chinese.
Reported scale: 1,700 firm Memories, more than 1 million custom workflows, weekly active users equal to 81% of monthly active users.
Reported customer outcome: one private equity firm reported $2.4 million in annualized savings from automated due diligence and investment memo workflows.