# BenchAudit > Independent audit of AI benchmarks scored on resistance to training-data contamination and grading leaks, plus the BenchAudit True Index for frontier models. Updated 2026-11-22 (Ref 2026-11). ## Core Pages - [BenchAudit Index](https://benchaudit.org/): Interactive and pre-rendered contamination ledger + True Index. - [Methodology](https://benchaudit.org/methodology/): Four-factor evaluation criteria (Privacy, Originality, Freshness, Grading) and Tiers S-D. - [Open Dataset (CC BY 4.0)](https://benchaudit.org/dataset/): Public Hugging Face dataset (bofinb/benchaudit) + BibTeX citation. - [About & Independence](https://benchaudit.org/about/): Independence statement, corrections policy, and contact (admin@benchaudit.org). - [Complete Markdown Index](https://benchaudit.org/md/index.md): Single-file Markdown summary of audited benchmarks and ranked models. - [Full LLM Context](https://benchaudit.org/llms-full.txt): Complete plain-Markdown snapshot of audited benchmarks and active frontier models. ## Open Dataset on Hugging Face (CC BY 4.0) - [BenchAudit on Hugging Face Hub (bofinb/benchaudit)](https://huggingface.co/datasets/bofinb/benchaudit) ## Benchmark Audits (Markdown Mirrors) - [AA-Briefcase v1.1 (Tier S, 10/10)](https://benchaudit.org/md/benchmarks/aa-briefcase-v1-1.md): Artificial Analysis · Knowledge Work - [AutomationBench-AA (Tier S, 10/10)](https://benchaudit.org/md/benchmarks/automationbench-aa.md): Artificial Analysis / Zapier · Knowledge Work - [FrontierMath (Tier S, 10/10)](https://benchaudit.org/md/benchmarks/frontiermath.md): Epoch AI · Science - [Harvey LAB-AA v1.1 (Tier S, 10/10)](https://benchaudit.org/md/benchmarks/harvey-lab-aa-v1-1.md): Artificial Analysis / Harvey · Knowledge Work - [MLCR-AA (Tier S, 10/10)](https://benchaudit.org/md/benchmarks/mlcr-aa.md): Artificial Analysis / Wisedocs · Science - [Vibe Code Bench 1-100 v1.1 (Tier S, 9.6/10)](https://benchaudit.org/md/benchmarks/vcb-1-100-v1-1.md): Vals AI · Coding - [FrontierCode-AA 2.0 (Tier S, 9.3/10)](https://benchaudit.org/md/benchmarks/frontiercode-aa-2-0.md): Artificial Analysis / Cognition · Coding - [MKA (Multimodal Knowledge Agent) (Tier S, 9/10)](https://benchaudit.org/md/benchmarks/mka.md): Artificial Analysis / Google DeepMind · Knowledge Work - [Vending-Bench 2 (Tier S, 9/10)](https://benchaudit.org/md/benchmarks/vending-bench-2.md): Andon Labs · Knowledge Work - [Code2Video-AA (Tier S, 8.8/10)](https://benchaudit.org/md/benchmarks/code2video-aa.md): Artificial Analysis / HeyGen · Coding - [EnterpriseOps-Gym-AA (Tier S, 8.8/10)](https://benchaudit.org/md/benchmarks/enterpriseops-gym-aa.md): Artificial Analysis / ServiceNow · Knowledge Work - [ITBench-AA (Tier S, 8.8/10)](https://benchaudit.org/md/benchmarks/itbench-aa.md): Artificial Analysis / IBM · Coding - [AA-AnalystAgent (Tier A, 8.3/10)](https://benchaudit.org/md/benchmarks/aa-analystagent.md): Artificial Analysis · Knowledge Work - [AA-Omniscience v1.1 (Tier A, 8.3/10)](https://benchaudit.org/md/benchmarks/aa-omniscience-v1-1.md): Artificial Analysis · General - [LiveBench (Tier A, 8.3/10)](https://benchaudit.org/md/benchmarks/livebench.md): LiveBench / Abacus.AI · General - [ARC-AGI-2 (Tier A, 8.2/10)](https://benchaudit.org/md/benchmarks/arc-agi-2.md): ARC Prize Foundation · General - [CritPt v2 (Tier A, 8/10)](https://benchaudit.org/md/benchmarks/critpt-v2.md): Artificial Analysis / CritPt · Science - [AA-3D (Tier A, 7.8/10)](https://benchaudit.org/md/benchmarks/aa-3d.md): Artificial Analysis · Coding - [Code Migration v1.1 (Tier A, 7.5/10)](https://benchaudit.org/md/benchmarks/code-migration-v1-1.md): Vals AI · Coding - [CyberBench v1.1 (Tier A, 7.5/10)](https://benchaudit.org/md/benchmarks/cyberbench.md): Vals AI · Coding - [SWE-bench Pro (Tier A, 7.5/10)](https://benchaudit.org/md/benchmarks/swe-bench-pro.md): Scale AI · Coding - [Agents' Last Exam v1.1 (Tier A, 7/10)](https://benchaudit.org/md/benchmarks/agents-last-exam-v1-1.md): UC Berkeley RDI · Knowledge Work - [HLE-Diamond (Tier B, 5.8/10)](https://benchaudit.org/md/benchmarks/hle-diamond.md): Artificial Analysis / CAIS / Scale AI · General - [GAIA (Tier B, 5.6/10)](https://benchaudit.org/md/benchmarks/gaia.md): Hugging Face / Meta / AutoGPT · General - [APEX-Agents-AA (Tier B, 5.5/10)](https://benchaudit.org/md/benchmarks/apex-agents-aa.md): Artificial Analysis / Mercor · Knowledge Work - [VIALS-AA (Tier B, 5.4/10)](https://benchaudit.org/md/benchmarks/vials-aa.md): Artificial Analysis / Handshake AI · Science - [OSWorld-2.1 (Offline subset) (Tier B, 5.2/10)](https://benchaudit.org/md/benchmarks/osworld-2-1-offline.md): OSWorld / XLANG Lab (HKU) · Knowledge Work - [Autoform-Hard (Tier C, 4.7/10)](https://benchaudit.org/md/benchmarks/autoform-hard.md): Vals AI · Science - [SWE Atlas - Codebase QnA (Tier C, 4.7/10)](https://benchaudit.org/md/benchmarks/swe-atlas-qna.md): Scale AI · Coding - [BioMysteryBench (Tier C, 4.5/10)](https://benchaudit.org/md/benchmarks/biomysterybench.md): Vals AI · Science - [GPQA Diamond (Tier C, 4.4/10)](https://benchaudit.org/md/benchmarks/gpqa-diamond.md): Artificial Analysis / NYU · Science - [Terminal-Bench 4.0 (Tier C, 4.4/10)](https://benchaudit.org/md/benchmarks/terminal-bench-4-0.md): Artificial Analysis / Harbor · Coding - [Terminal-Bench 4.1 (Tier C, 4.4/10)](https://benchaudit.org/md/benchmarks/terminal-bench-4-1.md): Artificial Analysis / Harbor · Coding - [DeepSWE v1.1 (Tier C, 4.3/10)](https://benchaudit.org/md/benchmarks/deepswe.md): Datacurve · Coding - [HealthBench (Tier C, 4.3/10)](https://benchaudit.org/md/benchmarks/healthbench.md): Artificial Analysis / OpenAI · Science - [Terminal-Bench-Science 0.1 (Tier C, 4.2/10)](https://benchaudit.org/md/benchmarks/terminal-bench-science-0-1.md): Artificial Analysis / Stanford / Harbor · Science - [Chartography (Tier C, 4.1/10)](https://benchaudit.org/md/benchmarks/chartography.md): Surge AI · Knowledge Work - [SWE-bench Pro V2 (Tier C, 3.9/10)](https://benchaudit.org/md/benchmarks/swe-bench-pro-v2.md): Scale AI / Reflection · Coding - [SWE-sweep (Tier C, 3.7/10)](https://benchaudit.org/md/benchmarks/swe-sweep.md): Meta Superintelligence Labs · Coding - [GDP.pdf (Tier C, 3.6/10)](https://benchaudit.org/md/benchmarks/gdp-pdf.md): Artificial Analysis / Surge AI · Knowledge Work - [GDPval-AA v2.1 (Tier C, 3.3/10)](https://benchaudit.org/md/benchmarks/gdpval-aa-v2-1.md): Artificial Analysis / OpenAI · Knowledge Work - [IFBench (Tier C, 3.2/10)](https://benchaudit.org/md/benchmarks/ifbench.md): Artificial Analysis / AllenAI · General - [LiveCodeBench v6 (Tier C, 3.2/10)](https://benchaudit.org/md/benchmarks/livecodebench.md): LiveCodeBench / UC Berkeley · Coding - [MMMU-Pro (Tier D, 2.1/10)](https://benchaudit.org/md/benchmarks/mmmu-pro.md): Artificial Analysis / MMMU · General - [SWE-bench Verified (Tier D, 1.5/10)](https://benchaudit.org/md/benchmarks/swe-bench-verified.md): SWE-bench / Princeton / OpenAI · Coding - [MMLU-Pro (Tier D, 1.1/10)](https://benchaudit.org/md/benchmarks/mmlu-pro.md): TIGER-AI-Lab / Artificial Analysis · General - [Cybench (Tier D, 0.5/10)](https://benchaudit.org/md/benchmarks/cybench.md): Cybench / Stanford · Coding ## Frontier Model Profiles (Top 20 Markdown Mirrors) - [#1 GPT-6.2 Astra (96.9p Top Tier)](https://benchaudit.org/md/models/gpt-6-2-astra__max.md) - [#2 Claude Fable 5.2 (95.3p Top Tier)](https://benchaudit.org/md/models/claude-fable-5-2__max.md) - [#3 Gemini 4 Neon (93.1p Top Tier)](https://benchaudit.org/md/models/gemini-4-neon__xhigh.md) - [#4 Gemini 4 Carbon (84.6p Top Tier)](https://benchaudit.org/md/models/gemini-4-carbon__high.md) - [#5 GPT-6.2 Sol (83.8p Top Tier)](https://benchaudit.org/md/models/gpt-6-2-sol__max.md) - [#6 Claude Opus 5.5 (83.0p Top Tier)](https://benchaudit.org/md/models/claude-opus-5-5__max.md) - [#7 GPT-6 Astra (83.0p Top Tier)](https://benchaudit.org/md/models/gpt-6-astra__max.md) - [#8 Claude Sonnet 5.5 (78.3p Top Tier)](https://benchaudit.org/md/models/claude-sonnet-5-5__max.md) - [#9 Muse Wonder Full (78.0p Top Tier)](https://benchaudit.org/md/models/muse-wonder-full__max.md) - [#10 GPT-6.1 Sol (77.7p Top Tier)](https://benchaudit.org/md/models/gpt-6-1-sol__max.md) - [#11 Gemini 4 Hydrogen (76.7p Top Tier)](https://benchaudit.org/md/models/gemini-4-hydrogen__xhigh.md) - [#12 Claude Fable 5.1 (70.7p Top Tier)](https://benchaudit.org/md/models/claude-fable-5-1__max.md) - [#13 GPT-6.1 Terra (70.5p Top Tier)](https://benchaudit.org/md/models/gpt-6-1-terra__max.md) - [#14 Muse Wonder (68.1p Top Tier)](https://benchaudit.org/md/models/muse-wonder__max.md) - [#15 Gemini 4 Argon (67.9p Top Tier)](https://benchaudit.org/md/models/gemini-4-argon__high.md) - [#16 GPT-6 Sol (66.4p Top Tier)](https://benchaudit.org/md/models/gpt-6-sol__max.md) - [#17 Kimi K3.1 Pro (65.3p Top Tier)](https://benchaudit.org/md/models/kimi-k3-1-pro__max.md) - [#18 Qwen4 Max (61.5p Top Tier)](https://benchaudit.org/md/models/qwen4-max__xhigh.md) - [#19 DeepSeek V4.1 Flash 1017 (56.0p Top Tier)](https://benchaudit.org/md/models/deepseek-v4-1-flash-1017__max.md) - [#20 Claude Haiku 5.5 (55.3p Top Tier)](https://benchaudit.org/md/models/claude-haiku-5-5__max.md)