Terminal X uses GPT-6.1 to evaluate investment output, reaching 88.6% agreement with senior investors' judgments

Terminal X's published data shows its Eval Agent scores investment output item by item against a 54-criteria scoring system, using GPT-6.1 Sol as the model. Compared with senior investors' judgments, the agreement rate is 88.6%; in identifying content that deviates from fund mandates, it finds 25% more than existing methods. For investment teams managing fund mandates and compliance review, this means checks once done by senior investors reviewing documents one by one can become batch automated evaluation, lowering per-review labor cost and expanding coverage. Note that 88.6% agreement means roughly one in ten judgments still differs from senior investors, so automated scores still need human review in high-risk decisions; all figures are self-published by Terminal X, and the original text does not disclose evaluation sample size or the specific criteria.