# Epoch AI Benchmarking Hub

Epoch AI runs AI models itself on a set of hard tests (maths, science, coding) and publishes the results in one place, so you can compare models on tests the labs did not grade themselves.

Category: aggregators. Type: leaderboard. Status: active.
Metric: Share of FrontierMath Tiers 1-3 (v2, private set) maths problems solved (%) (higher is better).
Source: Epoch AI — 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 | Score (%) | Release date |
|---|---|---|---|---|
| 1 | claude-fable-5-2_max | Anthropic | 95.1 | 2026-11-09 |
| 2 | gpt-6.1-sol_max | OpenAI | 93.7 | 2026-09-29 |
| 3 | gpt-6-astra_max | OpenAI | 93.7 | 2026-09-03 |
| 4 | claude-opus-5-5_max | Anthropic | 91.2 | 2026-09-22 |
| 5 | gpt-6.1-terra_max | OpenAI | 90.9 | 2026-10-26 |
| 6 | gemini-4-carbon_high | Google DeepMind | 90.5 | 2026-11-10 |
| 7 | claude-fable-5-1_max | Anthropic | 90.2 | 2026-09-01 |
| 8 | gpt-6-sol_max | OpenAI | 89.8 | 2026-09-22 |
| 9 | gpt-5.6-sol_max | OpenAI | 89.1 | 2026-07-09 |
| 10 | claude-sonnet-5-5_max | Anthropic | 88.8 | 2026-09-28 |
| 11 | gpt-5.5-pro_xhigh | OpenAI | 87.7 | 2026-04-23 |
| 12 | claude-fable-5_max | Anthropic | 87.0 | 2026-06-09 |
| 13 | gemini-4-argon_high | Google DeepMind | 86.0 | 2026-10-13 |
| 14 | gpt-5.6-terra_max | OpenAI | 86.0 | 2026-07-09 |
| 15 | claude-opus-5_max | Anthropic | 85.6 | 2026-07-24 |
| 16 | gpt-5.5_xhigh | OpenAI | 85.3 | 2026-04-23 |
| 17 | gpt-5.4-pro-2026-03-05_xhigh | OpenAI | 82.5 | 2026-03-05 |
| 18 | muse-wonder-1.0_max | Meta AI | 82.1 | 2026-10-12 |
| 19 | gpt-5.6-luna_max | OpenAI | 82.1 | 2026-07-09 |
| 20 | claude-opus-4-8_max | Anthropic | 80.0 | 2026-05-28 |
| 21 | gpt-6-luna_max | OpenAI | 78.9 | 2026-09-22 |
| 22 | gpt-5.4-2026-03-05_xhigh | OpenAI | 78.6 | 2026-03-05 |
| 23 | kimi-k3.1-pro-together_max | Moonshot | 78.2 | 2026-10-23 |
| 24 | claude-haiku-5-5_max | Anthropic | 75.1 | 2026-10-07 |
| 25 | qwen3.8-max_xhigh | Alibaba | 74.7 | 2026-08-02 |

Notes: The Hub covers dozens of benchmarks; only one representative table is shown here (FrontierMath Tiers 1-3 v2, Epoch's 'best score across scorers', converted from 0-1 to %). Sorted by score; the file itself is not sorted. Data zip generated 2026-11-17.
