Output Loop Bench · Reasoning

Decision race: AI models compared

Every business rated P1–P4 in batches: speed, and agreement with the original research. 7 models tested on the same task.

Rate trade businesses P1–P4

Every model rates the same 544 trade businesses P1–P4 in batches of 50. The top row is the original research banding, shown as a reference rather than the truth (its scores are hidden from the models), with how often each model lands in the same band.

Open Decision race in the interactive Bench