# AA-3D Contamination Audit

> AA-3D scores 7.8/10 (Tier A, Low Risk) on BenchAudit as of November 2026, reflecting public prompts with private held-out answers and isolated offline third-party grading.

- **Canonical URL:** https://benchaudit.org/benchmarks/aa-3d/
- **Benchmark ID:** `aa-3d`
- **Hub:** Artificial Analysis
- **Category:** Coding
- **Domains:** Science, Humanities, Coding
- **Skills:** Coding, Agentic, Multimodal, Instruction Following, Reasoning
- **BenchAudit Score:** 7.8/10 (Tier A - Low Risk)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 5.5 | 40% |
| Originality | 10 | 25% |
| Freshness | 8 | 20% |
| Grading | 10 | 15% |

## Primary-Source Evidence

### Privacy
> "Quality Elo comes entirely from blind, pairwise preference votes cast by our recruited panel. In each comparison, a panelist sees two scenes built for the same task by two different models, interacts with both, and selects the one they prefer. Public votes cast in AA-3D Voting do not enter the rating calculation."
- Source: https://artificialanalysis.ai/methodology/aa-3d

### Originality
> "AA-3D v1.0 measures how well AI models build interactive 3D experiences. Each model receives a written brief and builds a single web page with three.js in an offline sandbox."
- Source: https://artificialanalysis.ai/evaluations/aa-3d

### Freshness
> "Elo from blind pairwise votes · 95% confidence intervals · Higher is better · Data as at 12 October 2026"
- Source: https://artificialanalysis.ai/articles/aa-3d

### Grading
> "We run every model in Stirrup, our open-source agent harness. Each task runs in its own sandbox with no network access."
- Source: https://artificialanalysis.ai/methodology/aa-3d

## Top Frontier Models on AA-3D

| Rank | Model | Creator | Percentile |
| ---: | --- | --- | ---: |
| #1 | [GPT-6.2 Astra](https://benchaudit.org/models/gpt-6-2-astra__max/) | OpenAI | 100.0p |
| #2 | [Claude Fable 5.2](https://benchaudit.org/models/claude-fable-5-2__max/) | Anthropic | 98.8p |
| #3 | [GPT-6.2 Sol](https://benchaudit.org/models/gpt-6-2-sol__max/) | OpenAI | 95.0p |
| #4 | [Claude Sonnet 5.5](https://benchaudit.org/models/claude-sonnet-5-5__max/) | Anthropic | 91.1p |
| #5 | [Claude Opus 5.5](https://benchaudit.org/models/claude-opus-5-5__max/) | Anthropic | 87.2p |
| #6 | [GPT-6 Astra](https://benchaudit.org/models/gpt-6-astra__max/) | OpenAI | 83.3p |
| #7 | [Gemini 4 Argon](https://benchaudit.org/models/gemini-4-argon__high/) | Google | 79.4p |
| #8 | [Claude Fable 5.1](https://benchaudit.org/models/claude-fable-5-1__max/) | Anthropic | 75.6p |
| #9 | [GPT-6 Sol](https://benchaudit.org/models/gpt-6-sol__max/) | OpenAI | 71.7p |
| #10 | [GPT-6.1 Sol](https://benchaudit.org/models/gpt-6-1-sol__max/) | OpenAI | 67.8p |
