# FrontierMath Contamination Audit

> FrontierMath scores 10/10 (Tier S, Fortified) on BenchAudit as of November 2026, reflecting a strictly private held-out test set and isolated offline third-party grading.

- **Canonical URL:** https://benchaudit.org/benchmarks/frontiermath/
- **Benchmark ID:** `frontiermath`
- **Hub:** Epoch AI
- **Category:** Science
- **Domains:** Math, Science
- **Skills:** Reasoning, Coding, Tool Use
- **BenchAudit Score:** 10/10 (Tier S - Fortified)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

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

## Primary-Source Evidence

### Privacy
> "We have made twelve problems public: ten from Tiers 1-3 and two from Tier 4. Unless stated otherwise, all the numbers on this hub correspond to evaluations on the private sets."
- Source: https://epoch.ai/benchmarks/frontiermath-tier-4-v2

### Originality
> "We introduce FrontierMath, a benchmark of hundreds of original, exceptionally challenging mathematics problems crafted and vetted by expert mathematicians."
- Source: https://arxiv.org/abs/2411.04872

### Freshness
> "On 2026-06-12, we released a major update, addressing errors in 42% of problems. Following this update, the full FrontierMath dataset consists of 338 problems. This is split into a base set of 295 problems, which we call Tiers 1-3, and an expansion set of 43 exceptionally difficult problems, which we call Tier 4."
- Source: https://epoch.ai/benchmarks/frontiermath-tier-4-v2

### Grading
> "For each FrontierMath question, the model needs to submit a Python function answer() that returns the answer. The answer is a Python object, often (although not always) an integer or a sympy object. Our implementation allows the model to reason and run Python code."
- Source: https://epoch.ai/benchmarks/frontiermath-tier-4-v2

## Top Frontier Models on FrontierMath

| Rank | Model | Creator | Percentile |
| ---: | --- | --- | ---: |
| #1 | [Claude Fable 5.2](https://benchaudit.org/models/claude-fable-5-2__max/) | Anthropic | 99.5p |
| #1 | [GPT-6.1 Sol](https://benchaudit.org/models/gpt-6-1-sol__max/) | OpenAI | 99.5p |
| #1 | [GPT-6.2 Astra](https://benchaudit.org/models/gpt-6-2-astra__max/) | OpenAI | 99.5p |
| #4 | [GPT-6 Astra](https://benchaudit.org/models/gpt-6-astra__max/) | OpenAI | 93.1p |
| #5 | [Gemini 4 Carbon](https://benchaudit.org/models/gemini-4-carbon__high/) | Google | 89.9p |
| #6 | [Claude Opus 5.5](https://benchaudit.org/models/claude-opus-5-5__max/) | Anthropic | 86.8p |
| #7 | [GPT-6.1 Terra](https://benchaudit.org/models/gpt-6-1-terra__max/) | OpenAI | 83.6p |
| #8 | [GPT-6 Sol](https://benchaudit.org/models/gpt-6-sol__max/) | OpenAI | 80.5p |
| #9 | [Claude Fable 5.1](https://benchaudit.org/models/claude-fable-5-1__max/) | Anthropic | 75.7p |
| #9 | [Gemini 4 Hydrogen](https://benchaudit.org/models/gemini-4-hydrogen__xhigh/) | Google | 75.7p |
