# GAIA Contamination Audit

> GAIA scores 5.6/10 (Tier B, Moderate Risk) on BenchAudit as of November 2026, reflecting public prompts with private held-out answers and self-reported or live-network evaluation.

- **Canonical URL:** https://benchaudit.org/benchmarks/gaia/
- **Benchmark ID:** `gaia`
- **Hub:** Hugging Face / Meta / AutoGPT
- **Category:** General
- **Domains:** Business, Science, Humanities
- **Skills:** Agentic, Tool Use, Reasoning, Multimodal
- **BenchAudit Score:** 5.6/10 (Tier B - Moderate Risk)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 5.5 | 40% |
| Originality | 10 | 25% |
| Freshness | 2 | 20% |
| Grading | 3.3 | 15% |

## Primary-Source Evidence

### Privacy
> "Each level is divided into a fully public dev set for validation, and a test set with private answers and metadata."
- Source: https://huggingface.co/datasets/gaia-benchmark/GAIA

### Originality
> "We introduce GAIA, a benchmark for General AI Assistants that, if solved, would represent a milestone in AI research. GAIA proposes real-world questions that require a set of fundamental abilities such as reasoning, multi-modality handling, web browsing, and generally tool-use proficiency."
- Source: https://arxiv.org/abs/2311.12983

### Freshness
> "arXiv:2311.12983 (cs) [Submitted on 21 Nov 2023]"
- Source: https://arxiv.org/abs/2311.12983

### Grading
> "GAIA is a benchmark which aims at evaluating next-generation LLMs (LLMs with augmented capabilities due to added tooling, efficient prompting, access to search, etc)."
- Source: https://huggingface.co/datasets/gaia-benchmark/GAIA
