News
OpenAI says Terminal X’s Eval Agent matched senior investor judgment 88.6% and found 25% more mandate deviations, based on vendor-reported results.
Terminal X said its Eval Agent, running with GPT-6.1 Sol, matched senior investor judgment in 88.6% of cases and caught 25% more deviations from a fund mandate, according to an OpenAI customer case study published on 20 Oct. 2026. The report is notable because it frames AI as a workflow control tool for asset managers rather than as a trading signal, but the results are vendor-reported and were not presented as an independent audit.
Evidence and context
OpenAI’s case study said Terminal X grades outputs against a 54-criterion rubric covering source accuracy, analytical coherence and mandate alignment. Each criterion receives a pass or fail, with failures generating an explanation and corrected text or a better source, which then feeds later workflow instructions and test cases, according to the material published on 20 Oct. 2026.
The same case study reported 88.6% agreement with senior reviewers across 500 judgments, versus 86.4% for the next-best model. It also said the system achieved about 25% better detection of mandate departures and roughly 30% lower cost per query. Those figures are attributable to OpenAI’s case study, and they should be treated as product-performance claims rather than independently verified investment outcomes.
Terminal X’s broader product line is described in its 23 Sep. 2026 user guide as including research agents, custom workflows, private data rooms and personalization. Background material supplied with the case study says the company’s Capital Memory system captures a firm’s investment process, prior decisions and house views in customer-owned agents deployed on-premise or in the customer’s cloud.
The source material also identifies Terminal X co-founders Hyun Hong, its CEO, and Ken Lee, its CTO. The supplied material does not provide independent performance verification, client-level implementation data or evidence of market impact beyond the vendor-published case study.
News
OpenAI says Terminal X’s Eval Agent matched senior investor judgment 88.6% and found 25% more mandate deviations, based on vendor-reported results.
Terminal X said its Eval Agent, running with GPT-6.1 Sol, matched senior investor judgment in 88.6% of cases and caught 25% more deviations from a fund mandate, according to an OpenAI customer case study published on 20 Oct. 2026. The report is notable because it frames AI as a workflow control tool for asset managers rather than as a trading signal, but the results are vendor-reported and were not presented as an independent audit.
Evidence and context
OpenAI’s case study said Terminal X grades outputs against a 54-criterion rubric covering source accuracy, analytical coherence and mandate alignment. Each criterion receives a pass or fail, with failures generating an explanation and corrected text or a better source, which then feeds later workflow instructions and test cases, according to the material published on 20 Oct. 2026.
The same case study reported 88.6% agreement with senior reviewers across 500 judgments, versus 86.4% for the next-best model. It also said the system achieved about 25% better detection of mandate departures and roughly 30% lower cost per query. Those figures are attributable to OpenAI’s case study, and they should be treated as product-performance claims rather than independently verified investment outcomes.
Terminal X’s broader product line is described in its 23 Sep. 2026 user guide as including research agents, custom workflows, private data rooms and personalization. Background material supplied with the case study says the company’s Capital Memory system captures a firm’s investment process, prior decisions and house views in customer-owned agents deployed on-premise or in the customer’s cloud.
The source material also identifies Terminal X co-founders Hyun Hong, its CEO, and Ken Lee, its CTO. The supplied material does not provide independent performance verification, client-level implementation data or evidence of market impact beyond the vendor-published case study.
