Trust / benchmark / 2026-10 / v1
How quickly can you check an AI output?
A repeatable, timed check of instructions and arithmetic. Accuracy comes first. This measures your review task, not an AI model ranking.
Human review benchmark: two fixed multiple-choice items, identical instructions, server-recorded elapsed time, maximum 30 minutes. First completion per account per UTC month counts. Device and network delays affect time; scores are not a general intelligence measure.
Benchmark · Question version 1. This question collects responses each UTC calendar month. 2 task items; first completed attempts count, with a 30-minute session limit.
Voluntary reader sample, not a representative population survey. Published results remain public. How collection and results work →
Checking participation…
On the record
No published collection yet. Monthly results will appear here after collection closes and the minimum response threshold is reached.
Read the methodology →