November 2026 · Independent humanizer benchmarks. Explore the evidence

HumanizeBench November 2026 dataset

Version 2026-11-v1 contains 33 source texts, 759 saved rewrites, and measurements for 23 open-weight and commercial model entries. The snapshot includes 5,309 numeric detector results, three provider exclusions, and one pending check across seven detector routes.

Download version 2026-11-v1 JSON

Data dictionary

samples
Source texts, categories and prompts, keyed by source ID.
rows
Saved rewrites and their source IDs, detector values, evidence, readability and meaning. Detector probabilities range from 0 to 1. A null value represents a provider exclusion, and error represents a pending detector result.
entries
Ranked entries and their aggregate metrics, with coverage and source artifact details where available.
profile and coverage
Scoring weights, measurement configuration and counts for this snapshot.

Provenance and citation

Source texts and historical commercial rewrites are attributed to AI Humanizer Benchmark under CC BY 4.0. Cite HumanizeBench, November 2026 dataset, version 2026-11-v1, and cite upstream material where used. See the methodology for the scoring profile and limitations.

Supplemental Originality.ai measurements record repeated inputs separately. The current benchmark API reflects subsequent updates.