November 2026 · Independent humanizer benchmarks. Explore the evidence

About HumanizeBench

Open weights. Closed services. Shared measurements.

HumanizeBench presents a November 2026 evaluation of thirteen open-weight repository entries and ten sets of published service rewrites. The site is a new presentation of our recorded evaluation, not the reference website’s rankings.

The source of the data

We preserve the original AI Humanizer Benchmark’s 33 source texts and scoring script. We generate 429 open-weight outputs and re-score all 759 outputs under the same agreed NVIDIA quality substitutes and current detector routes.

Current scope

This is one evaluation cycle. We do not have multi-month trend data, product pricing studies, or test videos. The site shows only what our saved records support.