Terminal X Eval Agent Matches Senior Investor Judgment 88.6% Using GPT-6.1 Sol

• Terminal X’s Eval Agent, powered by GPT-6.1 Sol, agrees with senior human reviewers on 88.6% of 500 scored judgments, outperforming the next-best model by 220 basis points.
• The agent detects 25% more recommendations that stray from a fund’s mandate and operates at approximately 30% lower cost per query than the next-best model.
• One private equity customer reported $2.4 million in annualized savings from automated due diligence and investment memo workflows.
• The system encodes firm-specific investment logic into 1,700 firm Memories, running over 1 million custom workflows with 81% of monthly active users engaging weekly.

Evaluation and Model Selection

• Eval Agent grades outputs against a 54-criterion rubric, checking numbers, sources, and alignment with house views.
• GPT-6.1 Sol handles high-judgment stages requiring large context windows, while GPT-6 Luna manages high-volume, token-efficient steps.
• Failures in evaluation generate new Memory entries, encoding corrections as workflow instructions or test cases.

Implementation and Scope

• Forward-deployed engineers reverse-engineer trusted memos and undocumented processes to build customized agents.
• Agents operate within private data rooms, processing documents in English, Korean, Japanese, and Chinese.
• Every production query is auditable with a full trail of steps, sources, models, and reasoning.