Skip to content

Benchmarks Methods Kit Data About

by Cheetah

Accuracy
Safe to hand off
Cost
Speed
Method and data

Benchmarks/AI judges

AI judges for data hygiene, Nov 2026

How we scored this

Correct means the model's first answer matched the answer we already knew. Each score comes with a likely range, because a few hundred records can't give an exact figure.

When our test can't separate a model from the leader, both are in the top tier. Too close to call is not the same as equal. We keep the answer keys hidden. Read the methods.

The four jobs

Models and repeats

How we find the top tier

What we did not test

Data and corrections

Methods for this report
Download all data (JSON)
Run it on your own records
Send corrections to [email protected].

Revenue Bench by Cheetah No vendor sponsorship and no affiliate links. Code is MIT licensed. Aggregates are CC-BY-4.0.