Frontier models score 1% exact-match on ICD-10 coding when the task is following a real manual.

TAM curates tasks from ICD-10-CM clinical coding and US federal sentencing where each answer requires executing interdependent steps across different sections of a rulebook containing tens of thousands of rules. RAG, ReAct prompting and an agent harness on GPT-5 all reach 1% on ICD-10-CM and 15.5% on sentencing. If your product plan involves an agent following a long procedural rulebook, this is the number to plan against.