Hugging Face Dataset (v1.0)

CC BY 4.0 · Periodic Hub releases

bofinb/benchaudit

The public BenchAudit dataset subset is distributed via the Hugging Face Hub with versioned commits, interactive Dataset Viewer exploration, and direct datasets library loading.

benchmarks config

BenchAudit Score (0-10), Tier (S-D), Privacy, Originality, Freshness, and Grading scores, structured states, refresh dates, and canonical audit URLs.

models config

True Index rank, Top-Tier (Tier S/A) average percentile, Low-Tier (Tier C/D) average percentile, and within-domain matched Low-Tier Boost.

from datasets import load_dataset

# Load audited benchmarks
benchmarks = load_dataset("bofinb/benchaudit", "benchmarks", split="train")

# Load frontier model True Index
models = load_dataset("bofinb/benchaudit", "models", split="train")

How to Cite BenchAudit

CC BY 4.0 License ↗

You may copy, redistribute, and adapt the public dataset subset for research, journalism, or model training under Creative Commons Attribution 4.0 International (CC BY 4.0) with attribution to BenchAudit (https://benchaudit.org/).

@dataset{benchaudit_2026_11,
  author    = {Bofin Babu},
  title     = {BenchAudit: AI Benchmark Contamination and Integrity Index (v1.0)},
  year      = {2026},
  publisher = {Hugging Face / BenchAudit},
  url       = {https://huggingface.co/datasets/bofinb/benchaudit},
  license   = {CC-BY-4.0}
}