Terminal X says its GPT-6.1 Sol-powered evaluation agent matched senior human reviewers 88.6% of the time across 500 investment judgments.
OpenAI described how Terminal X uses GPT-6.1 Sol and GPT-6 Luna in its Capital Memory platform, which builds AI workflows around the investment practices of individual firms. In tests reported by Terminal X, its evaluation agent using GPT-6.1 Sol agreed with senior human reviewers on 88.6% of 500 scored judgments. The next-best model reached 86.4%, a difference of 220 basis points.
The system checks an agent’s recommendations against a 54-criterion rubric covering matters such as source selection, financial figures, comparable companies, narrative coherence and consistency with a fund’s investment mandate. Terminal X says GPT-6.1 Sol helped the evaluator identify about 25% more recommendations that departed from the mandate than the next-best model. It also reports that the same workflows cost about 30% less per query, at similar speed.
Terminal X works with investment teams to capture practices found in internal memos and in how investors make decisions. Those processes are organized into customized agents and firm-specific “Memories,” which can include previous analyses, committee decisions and workflow instructions. The agents run in each customer’s private data environment, and production queries retain an audit trail of the steps, sources and model used.
The company says its systems support document processing in English, Korean, Japanese and Chinese. One private-equity customer reported $2.4 million in annualized savings from automated due diligence and investment-memo workflows. OpenAI’s post also reports that Terminal X had created 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.
Terminal X says it plans to extend the Memory approach beyond investment desks to individual professionals. The reported comparisons are the company’s evaluation results; they describe performance on its scored investment judgments, not a general measure of model performance across finance.