Skip to main content
BenchLM
Data

FrontierMath v2 Tier 4 (FrontierMath v2 (Tier 4))

Data verified

Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning.

Top models on FrontierMath v2 (Tier 4) — October 29, 2026

As of October 29, 2026, GPT-6.1 Sol leads the FrontierMath v2 (Tier 4) leaderboard with 100.000% , followed by GPT-6 Astra (97.600%) and Claude Opus 5.5 (95.000%).

1Closed

GPT-6.1 Sol

OpenAI

100.000%
Overall 85.27Context 1.05M
2Closed

GPT-6 Astra

OpenAI

97.600%
Overall 86.07Context 1.05M
3Closed

Claude Opus 5.5

Anthropic

95.000%
Overall 86.29Context 1M

55 modelsMathematics10% of category scoreCurrentUpdated October 29, 2026

Leaderboard (55 models)

FrontierMath v2 (Tier 4) leaderboard
RankModel / configurationScoreParameters (B)Open / closed
1GPT-6.1 SolOpenAI
100.000%
Not reportedClosed
2GPT-6 AstraOpenAI
97.600%
Not reportedClosed
3Claude Opus 5.5Anthropic
95.000%
Not reportedClosed
4GPT-6.1 TerraOpenAI
92.683%
Not reportedClosed
5Claude Fable 5Anthropic
90.200%
Not reportedClosed
6GPT-6 SolOpenAI
90.000%
Not reportedClosed
7Claude Fable 5.1Anthropic
87.805%
Not reportedClosed
8GPT-5.6 SolOpenAI
83.000%
Not reportedClosed
9Claude Sonnet 5.5Anthropic
80.488%
Not reportedClosed
10Gemini 4 ArgonGoogle
78.049%
Not reportedClosed
11GPT-5.5 ProOpenAI
78.049%
Not reportedClosed
12Claude Opus 5Anthropic
73.171%
Not reportedClosed
13GPT-5.5OpenAI
72.500%
Not reportedClosed
14GPT-5.6 TerraOpenAI
68.300%
Not reportedClosed
15GPT-5.4 ProOpenAI
58.537%
Not reportedClosed
16GPT-5.6 LunaOpenAI
58.500%
Not reportedClosed
17Claude Opus 4.8Anthropic
56.098%
Not reportedClosed
18GPT-6 LunaOpenAI
56.098%
Not reportedClosed
19GPT-5.4OpenAI
49.000%
Not reportedClosed
20Qwen3.8 MaxAlibaba
46.341%
Not reportedOpen
21Claude Haiku 5.5Anthropic
46.341%
Not reportedClosed
22Muse Spark 1.3Meta
46.341%
Not reportedClosed
23Gemini 3.7 FlashGoogle
36.585%
Not reportedClosed
24Qwen3.7 MaxAlibaba
34.146%
Not reportedClosed
25Grok 4.6xAI
31.707%
Not reportedClosed
26Claude Opus 4.7Anthropic
31.707%
Not reportedClosed
27GPT-5.2OpenAI
31.700%
Not reportedClosed
28Claude Sonnet 5Anthropic
29.268%
Not reportedClosed
29GLM-5.2Z.AI
29.268%
Not reportedOpen
30GLM-5.3Z.AI
29.268%
Not reportedOpen
31Gemini 3.1 ProGoogle
26.829%
Not reportedClosed
32Gemini 3.5 FlashGoogle
26.829%
Not reportedClosed
33Claude Opus 4.6Anthropic
26.829%
Not reportedClosed
34DeepSeek V4.1 Flash 1017DeepSeek
26.829%
Not reportedOpen
35Kimi K2.6Moonshot AI
25.641%
Not reportedOpen
36Grok 4.5xAI
24.390%
Not reportedClosed
37Gemini 3.8 FlashGoogle
21.951%
Not reportedClosed
38Gemini 3.6 FlashGoogle
21.951%
Not reportedClosed
39Inkling-SmallThinking Machines Lab
17.073%
Not reportedOpen
40Gemini 3 FlashGoogle
17.073%
Not reportedClosed
41GLM-5.3-FlashZ.AI
17.073%
Not reportedOpen
42Grok 4.7xAI
17.073%
Not reportedClosed
43Grok 4.3xAI
14.634%
Not reportedClosed
44Kimi K2.7 CodeMoonshot AI
12.195%
Not reportedOpen
45GPT-5.4 nanoOpenAI
12.195%
Not reportedClosed
46GPT-5 miniOpenAI
12.195%
Not reportedClosed
47GPT-5.4 miniOpenAI
9.756%
Not reportedClosed
48InklingThinking Machines Lab
4.878%
Not reportedOpen
49Claude Opus 4.5Anthropic
4.878%
Not reportedClosed
50o4-mini (high)OpenAI
4.878%
Not reportedClosed
51Claude Sonnet 4.5Anthropic
2.439%
Not reportedClosed
52DeepSeek V4 ProDeepSeek
2.439%
Not reportedOpen
53GPT-5 nanoOpenAI
2.439%
Not reportedClosed
54Gemini 2.5 ProGoogle
0.000%
Not reportedClosed
55Gemini 3.5 Flash-LiteGoogle
0.000%
Not reportedClosed

According to BenchLM.ai, GPT-6.1 Sol leads the FrontierMath v2 (Tier 4) benchmark with a score of 100.000%, followed by GPT-6 Astra (97.600%) and Claude Opus 5.5 (95.000%). There is significant spread across the leaderboard, making this benchmark effective at differentiating model capabilities.

55 models have been evaluated on FrontierMath v2 (Tier 4). The benchmark falls in the Mathematics category. The Mathematics leaderboard ranks models by a weighted category score, and FrontierMath v2 (Tier 4) contributes 10% of it. BenchAlign v5.8 also uses it in the overall ranking.

About FrontierMath v2 (Tier 4)

Year

2026

Tasks

41 private extreme-difficulty mathematics problems

Format

Python-enabled iterative mathematical problem solving

Difficulty

Research-level mathematics requiring hours or days of expert work

FrontierMath v2 Tier 4 contains 43 problems, including two public examples. Published scores use the 41 private problems. Its smaller sample makes individual scores noisier than Tiers 1-3, so it is a separate ranking factor with lower weight rather than being averaged into the core score.

FrontierMath Tier 4 v2 leaderboard

Freshness and provenance

Version

FrontierMath v2 (Tier 4) 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

Current

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does FrontierMath v2 (Tier 4) measure?

Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning.

Which model scores highest on FrontierMath v2 (Tier 4)?

GPT-6.1 Sol by OpenAI currently leads with a score of 100.000% on FrontierMath v2 (Tier 4).

How many models are evaluated on FrontierMath v2 (Tier 4)?

55 AI models have published results on FrontierMath v2 (Tier 4) in the BenchLM catalog.

Compare top models on FrontierMath v2 (Tier 4)

GPT-6.1 Sol vs GPT-6 AstraGPT-6 Astra vs Claude Opus 5.5Claude Opus 5.5 vs GPT-6.1 TerraGPT-6.1 Terra vs Claude Fable 5
Last updated: October 29, 2026 · BenchLM version FrontierMath v2 (Tier 4) 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 6,100+ subscribers.

One email each week. Unsubscribe anytime.