FrontierMath Tier 4 v2 is a private expansion set of 43 exceptionally difficult research-level mathematics problems. Released on June 12, 2026, v2 corrected 12 Tier 4 problems and removed seven.
Browse the latest scores, model modes, release dates, and parameter sizes for FrontierMath Tier 4 v2.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
| Rank | Model | License | |||
|---|---|---|---|---|---|
 Claude Fable 5 Thinking Level · Max | 87.80 | 2026-06-09 | Unknown | Closed | |
 GPT-5.6 Sol Thinking Level · Max | 82.93 | 2026-06-26 | Unknown | Closed | |
GPT-5.5 Pro Thinking Level · Extra High | 78.05 | 2026-04-23 | Unknown | Closed | |
4 | Claude Opus 5 Thinking Level · Max | 73.17 | 2026-07-24 | Unknown | Closed |
5 | GPT-5.5 Thinking Level · Extra High | 72.50 | 2026-04-23 | Unknown | Closed |
6 | GPT-5.6 Terra Thinking Level · Max | 70.73 | 2026-06-26 | Unknown | Closed |
7 | GPT-5.6 Luna Thinking Level · Max | 60.98 | 2026-06-26 | Unknown | Closed |
8 | GPT-5.4 Pro Thinking Level · Extra High | 58.54 | 2026-03-05 | Unknown | Closed |
9 | Claude Opus 4.8 Thinking Level · Max | 56.10 | 2026-05-28 | Unknown | Closed |
10 | GPT-5.4 Thinking Level · Extra High | 49.00 | 2026-03-05 | Unknown | Closed |
11 |  Qwen3.8-Max Thinking Level · Extra High | 46.34 | 2026-08-03 | 2400B | Closed |
12 | GPT-5.2 Pro Thinking Level · Extra High | 46.00 | 2025-12-11 | Unknown | Closed |
13 |  Kimi K3 Thinking Level · Max | 39.02 | 2026-07-16 | 2800B | Free Commercial |
14 | Opus 4.7 Thinking Level · Max | 31.71 | 2026-04-16 | Unknown | Closed |
15 | GPT-5.2 Thinking Level · Extra High | 31.70 | 2025-12-11 | Unknown | Closed |
16 | Claude Sonnet 5 Thinking Level · Max | 29.27 | 2026-06-30 | Unknown | Closed |
17 |  GLM-5.2 Thinking Level · Max | 29.27 | 2026-06-13 | 753.3B | Free Commercial |
18 | Claude Opus 4.6 Thinking Level · Max | 26.83 | 2026-02-05 | Unknown | Closed |
19 |  Gemini 3.5 Flash Thinking Level · High | 26.83 | 2026-06-20 | Unknown | Closed |
20 |  Grok 4.5 Thinking Level · High | 24.39 | 2026-07-08 | Unknown | Closed |
21 | GPT-5 Thinking Level · High | 21.95 | 2025-08-07 | Unknown | Closed |
22 | Gemini 3.6 Flash Thinking Level · High | 21.95 | 2026-07-21 | Unknown | Closed |
23 | GPT-5-Pro Thinking Level · High | 19.51 | 2025-08-07 | Unknown | Closed |
24 | Grok 4.3 Beta Thinking Level · High | 14.63 | 2026-05-17 | 500B | Closed |
25 | GPT-5-mini Thinking Level · High | 12.20 | 2025-08-07 | Unknown | Closed |
26 | GPT-5.4 nano Thinking Level · High | 12.20 | 2026-03-17 | Unknown | Closed |
27 | GPT-5.4 mini Thinking Level · Extra High | 9.76 | 2026-03-17 | Unknown | Closed |
28 | OpenAI o4 - mini Thinking Level · High | 4.88 | 2025-04-16 | Unknown | Closed |
29 | Opus 4.5 32K | 4.88 | 2025-11-25 | Unknown | Closed |
30 | Opus 4.1 32K | 2.44 | 2025-08-06 | Unknown | Closed |
In Epoch AI's data observed August 12, 2026, Claude Fable 5 (max) scores 87.8, GPT-5.6 Sol (max) 82.9, and GPT-5.5 Pro (xhigh) 78.0.
