EarlyEval Halts Agent Runs Early, Cuts Input Tokens 44% at 89–97% Accuracy

Agent evaluation cost has become a genuine bottleneck to iterative development; a method that cuts token spend 44% with near-zero impact on measured resolve rates is immediately practical for any team running agentic benchmarks.