Benchmark

FrontierMath Tier 4 (v1)

Measures: The first private set of 48 exceptionally difficult research-level math problems in FrontierMath's Tier 4; replaced by v2 on 2026-06-12.

Maintainer
Epoch AI
Introduced
2025
Status
Superseded
Saturation
Headroom
Typical provenance
Independent
In scoring set
No

PaperOfficial leaderboard

Leaderboard

Independent and third-party results only.

Leaderboard
ModelAccuracyProvenanceSourceEvaluated
Claude Opus 4.8
Anthropic · Max effort
31.3%IndependentEpoch AI8 Jun 2026
GPT-5.4
OpenAI · Xhigh effort
27.1%IndependentEpoch AI6 Mar 2026
Claude Opus 4.7
Anthropic · Xhigh effort
22.9%IndependentEpoch AI17 Apr 2026
Claude Opus 4.6
Anthropic · Max effort
22.9%IndependentEpoch AI12 Feb 2026
Claude Opus 4.6
Anthropic · 64K effort
20.8%IndependentEpoch AI6 Feb 2026
Claude Opus 4.6
Anthropic · 32K effort
20.8%IndependentEpoch AI6 Feb 2026
GPT-5.2
OpenAI · Xhigh effort
18.8%IndependentEpoch AI14 Dec 2025
GPT-5.2
OpenAI · High effort
18.8%IndependentEpoch AI11 Dec 2025
Gemini 3 Pro18.8%IndependentEpoch AI21 Nov 2025
Gemini 3.1 Pro16.7%IndependentEpoch AI19 Feb 2026
GPT-5.2
OpenAI · Medium effort
16.7%IndependentEpoch AI11 Dec 2025
GPT-5 Pro
OpenAI · High effort
14.6%IndependentEpoch AI9 Oct 2025
Claude Opus 4.614.6%IndependentEpoch AI5 Feb 2026
Gemini 3.5 Flash
Google DeepMind · High effort
14.6%IndependentEpoch AI25 May 2026
Kimi K2.614.6%IndependentEpoch AI8 May 2026
GLM-5.112.5%IndependentEpoch AI12 May 2026
GPT-5
OpenAI · High effort
12.5%IndependentEpoch AI30 Oct 2025
GPT-5.1
OpenAI · High effort
12.5%IndependentEpoch AI13 Nov 2025
Qwen3.6-Plus8.3%IndependentEpoch AI12 May 2026
Claude Sonnet 4.6
Anthropic · 16K effort
8.3%IndependentEpoch AI21 Feb 2026
GPT-5
OpenAI · Medium effort
6.3%IndependentEpoch AI7 Aug 2025
GPT-5 mini
OpenAI · High effort
6.3%IndependentEpoch AI30 Oct 2025
GPT-5.2
OpenAI · Low effort
6.3%IndependentEpoch AI11 Dec 2025
GPT-5.4 nano
OpenAI · High effort
6.3%IndependentEpoch AI15 Apr 2026
o4-mini
OpenAI · High effort
6.3%IndependentEpoch AI1 Jul 2025
Claude Opus 4
Anthropic · 27K effort
4.2%IndependentEpoch AI1 Jul 2025
Claude Opus 4.1
Anthropic · 27K effort
4.2%IndependentEpoch AI5 Aug 2025
Claude Opus 4.54.2%IndependentEpoch AI25 Nov 2025
Claude Opus 4.5
Anthropic · 32K effort
4.2%IndependentEpoch AI25 Nov 2025
Claude Sonnet 4.5
Anthropic · 32K effort
4.2%IndependentEpoch AI22 Oct 2025
Gemini 2.5 Flash4.2%IndependentEpoch AI18 Dec 2025
Gemini 2.5 Pro4.2%IndependentEpoch AI3 Jul 2025
Gemini 3 Flash4.2%IndependentEpoch AI17 Dec 2025
GPT-5 mini
OpenAI · Medium effort
4.2%IndependentEpoch AI7 Aug 2025
GPT-5.1
OpenAI · Medium effort
4.2%IndependentEpoch AI17 Nov 2025
o3-mini
OpenAI · High effort
4.2%IndependentEpoch AI1 Jul 2025
Qwen3.6-Max-Preview4.2%IndependentEpoch AI28 May 2026
GLM-4.62.1%IndependentEpoch AI8 Dec 2025
GLM-52.1%IndependentEpoch AI19 Feb 2026
Claude Haiku 4.5
Anthropic · 32K effort
2.1%IndependentEpoch AI22 Oct 2025
Claude Opus 4.5
Anthropic · 16K effort
2.1%IndependentEpoch AI25 Nov 2025
Claude Sonnet 4.52.1%IndependentEpoch AI29 Sep 2025
Gemini 2.5 Pro2.1%IndependentEpoch AI3 Jul 2025
GPT-5 nano
OpenAI · Medium effort
2.1%IndependentEpoch AI7 Aug 2025
Grok 42.1%IndependentEpoch AI11 Aug 2025
o3
OpenAI · High effort
2.1%IndependentEpoch AI1 Jul 2025
o4-mini
OpenAI · Medium effort
2.1%IndependentEpoch AI7 Aug 2025
Qwen3.5-Plus2.1%IndependentEpoch AI15 May 2026
GPT-5.4 mini
OpenAI · High effort
2.1%IndependentEpoch AI15 Apr 2026
Claude 3.5 Sonnet0.0%IndependentEpoch AI1 Jul 2025
Claude 3.5 Sonnet0.0%IndependentEpoch AI1 Jul 2025
Claude Opus 40.0%IndependentEpoch AI1 Jul 2025
Claude Sonnet 40.0%IndependentEpoch AI1 Jul 2025
GLM-4.70.0%IndependentEpoch AI30 Jan 2026
GPT-4.10.0%IndependentEpoch AI1 Jul 2025
GPT-5 nano
OpenAI · High effort
0.0%IndependentEpoch AI30 Oct 2025
Grok 30.0%IndependentEpoch AI1 Jul 2025
Kimi K2 Thinking0.0%IndependentEpoch AI4 Dec 2025
Qwen3-235B-A22B-Thinking-25070.0%IndependentEpoch AI11 Dec 2025
Qwen3.5-Flash0.0%IndependentEpoch AI12 May 2026
Qwen3.6-Flash0.0%IndependentEpoch AI12 May 2026
  • Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0.

Sources