# IFBench Contamination Audit

> IFBench scores 3.2/10 (Tier C, High Risk) on BenchAudit as of November 2026, reflecting publicly downloadable plaintext test files and standardized independent third-party grading.

- **Canonical URL:** https://benchaudit.org/benchmarks/ifbench/
- **Benchmark ID:** `ifbench`
- **Hub:** Artificial Analysis / AllenAI
- **Category:** General
- **Domains:** Coding, Humanities
- **Skills:** Instruction Following, Writing, Reasoning
- **BenchAudit Score:** 3.2/10 (Tier C - High Risk)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 0 | 40% |
| Originality | 7.2 | 25% |
| Freshness | 1.5 | 20% |
| Grading | 7.3 | 15% |

## Primary-Source Evidence

### Privacy
> "Uses the single-turn IFBench dataset, which contains 294 questions"
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking

### Originality
> "OOD Constraints: 58 new and challenging constraints, with corresponding verification functions. The constraint templates are combined with prompts from a held-out set of WildChat (Zhao et al. 2024)."
- Source: https://github.com/allenai/IFBench

### Freshness
> "arXiv:2507.02833 (cs) [Submitted on 3 Jul 2025 ( v1 ), last revised 11 Nov 2025 (this version, v3)]"
- Source: https://arxiv.org/abs/2507.02833

### Grading
> "We employ the loose evaluation mode to robustly assess instruction-following, which accounts for extraneous text or formatting by checking several variations of the model's output (e.g., with and without the first and last lines, and with asterisks removed)"
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking
