# FrontierMath

Hundreds of brand-new, unpublished math problems written by professional mathematicians, ranging from hard competition level up to problems that take experts days. Because the problems are secret, it is a strong check of real mathematical ability rather than memorization.

Category: math. Type: leaderboard. Status: active.
Metric: Share of Tiers 1-3 problems solved correctly (%) (higher is better).
Source: Epoch AI (Benchmarking Hub) — https://epoch.ai/data/benchmark_data.zip, results as of 2026-11-17, retrieved 2026-11-17, license: CC-BY-4.0.

| Rank | Model | Organization | Tiers 1-3 (%) | Tier 4 (%) | Release date |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 (max) | Anthropic | 95.1 | 100.0 | 2026-11-09 |
| 2 | GPT-6.1 Sol (max) | OpenAI | 93.7 | 100.0 | 2026-09-29 |
| 3 | GPT-6 Astra (max) | OpenAI | 93.7 | 97.6 | 2026-09-03 |
| 4 | Claude Opus 5.5 (max) | Anthropic | 91.2 | 95.0 | 2026-09-22 |
| 5 | GPT-6.1 Terra (max) | OpenAI | 90.9 | 92.7 | 2026-10-26 |
| 6 | Gemini 4 Carbon (high) | Google DeepMind | 90.5 | 95.1 | 2026-11-10 |
| 7 | Claude Fable 5.1 (max) | Anthropic | 90.2 | 87.8 | 2026-09-01 |
| 8 | GPT-6 Sol (max) | OpenAI | 89.8 | 90.0 | 2026-09-22 |
| 9 | GPT-5.6 Sol (max) | OpenAI | 89.1 | 82.9 | 2026-07-09 |
| 10 | Claude Sonnet 5.5 (max) | Anthropic | 88.8 | 80.5 | 2026-09-28 |
| 11 | GPT-5.5 Pro (xhigh) | OpenAI | 87.7 | 78.0 | 2026-04-23 |
| 12 | Claude Fable 5 (max) | Anthropic | 87.0 | 90.2 | 2026-06-09 |
| 13 | Gemini 4 Argon (high) | Google DeepMind | 86.0 | 78.0 | 2026-10-13 |
| 14 | GPT-5.6 Terra (max) | OpenAI | 86.0 | 70.7 | 2026-07-09 |
| 15 | Claude Opus 5 (max) | Anthropic | 85.6 | 73.2 | 2026-07-24 |
| 16 | GPT-5.5 (xhigh) | OpenAI | 85.3 | 72.5 | 2026-04-23 |
| 17 | GPT-5.4 Pro (xhigh) | OpenAI | 82.5 | 58.5 | 2026-03-05 |
| 18 | Muse Wonder (max) | Meta AI | 82.1 | 70.7 | 2026-10-12 |
| 19 | GPT-5.6 Luna (max) | OpenAI | 82.1 | 61.0 | 2026-07-09 |
| 20 | Claude Opus 4.8 | Anthropic | 80.0 | 56.1 | 2026-05-28 |
| 21 | GPT-6 Luna (max) | OpenAI | 78.9 | 56.1 | 2026-09-22 |
| 22 | GPT-5.4 (xhigh) | OpenAI | 78.6 | 49.0 | 2026-03-05 |
| 23 | Kimi K3.1 Pro (Together, max) | Moonshot | 78.2 | 65.9 | 2026-10-23 |
| 24 | Claude Haiku 5.5 (max) | Anthropic | 75.1 | 46.3 | 2026-10-07 |
| 25 | Qwen3.8 Max (xhigh) | Alibaba | 74.7 | 46.3 | 2026-08-02 |

Notes: Epoch AI runs on the private v2 problem sets (frontiermath_tiers_1_3_v2.csv and frontiermath_tier_4_v2.csv), converted to % (1 decimal). Tier 4 is the research-level tier; blank = not evaluated on Tier 4 v2. OpenAI funded the benchmark and has access to much of the problem set.
