Skip to content

Benchmark

FrontierMath Tier 4 (v2)

Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.

Models measured
74
Scores recorded
104
Sources
3
IQ weight
2.74

Score distribution

14.5%
Mean
0.561%
Median
29.1%
Std dev
0.049%
10th percentile
72.5%
90th percentile
0%
Min
0%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.74 for chat models, measured over 74 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51GPT-5.2 Proxhigh· Epoch AI Benchmarking Hub data46%
52Muse Spark 1.3xhigh· Epoch AI Benchmarking Hub data41.5%
53GPT-5.5 Prov2 · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.39.6%
54Kimi K3max· Epoch AI Benchmarking Hub data39%
55GPT-5.4 ProOpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.38%
56Gemini 3.7 Flashhigh· Epoch AI Benchmarking Hub data36.6%
57GPT 5.5v2 · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.35.4%
58Qwen3.7 Max· Epoch AI Benchmarking Hub data34.1%
59Qwen3.8 Maxxhigh· Epoch AI Benchmarking Hub data34.1%
60Claude Opus 4.7max· Epoch AI Benchmarking Hub data31.7%
61Grok 4.6xhigh· Epoch AI Benchmarking Hub data31.7%
62Muse Spark Open 816B A42Bxhigh· Epoch AI Benchmarking Hub data31.7%
63GPT 5.2xhigh· Epoch AI Benchmarking Hub data31.7%
64GPT-5.2 ProOpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31.3%
65Claude Sonnet 5max· Epoch AI Benchmarking Hub data29.3%
66GLM 5.2max· Epoch AI Benchmarking Hub data29.3%
67GLM 5.3max· Epoch AI Benchmarking Hub data29.3%
68GPT 5.4OpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.27.1%
69Claude Opus 4.6max· Epoch AI Benchmarking Hub data26.8%
70DeepSeek V4 Pro 0813max· Epoch AI Benchmarking Hub data26.8%
71DeepSeek V4.1 Flashmax· Epoch AI Benchmarking Hub data26.8%
72Gemini 3.1 Pro Preview· Epoch AI Benchmarking Hub data26.8%
73Gemini 3.5 Flashhigh· Epoch AI Benchmarking Hub data26.8%
74Mistral Large 4high· Epoch AI Benchmarking Hub data26.8%
75Kimi K2.6· Epoch AI Benchmarking Hub data25.6%
76DeepSeek V4 Flash 0731max· Epoch AI Benchmarking Hub data24.4%
77Grok 4.5high· Epoch AI Benchmarking Hub data24.4%
78Gemini 3.6 Flashhigh· Epoch AI Benchmarking Hub data22%
79Gemini 3.8 Flashhigh· Epoch AI Benchmarking Hub data22%
80GPT 5high· Epoch AI Benchmarking Hub data22%
81GPT-5 Prohigh· Epoch AI Benchmarking Hub data19.5%
82GPT 5.2OpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.18.8%
83Gemini 3 Flash· Epoch AI Benchmarking Hub data17.1%
84GLM 5.3 Flashmax· Epoch AI Benchmarking Hub data17.1%
85Grok 4.20reasoning, 0309· Epoch AI Benchmarking Hub data17.1%
86Grok 4.7xhigh· Epoch AI Benchmarking Hub data17.1%
87Inkling Smallxhigh· Epoch AI Benchmarking Hub data17.1%
88Grok 4.3high· Epoch AI Benchmarking Hub data14.6%
89GPT 5.2Python · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.14.6%
90GPT 5.4 Nanohigh· Epoch AI Benchmarking Hub data12.2%
91GPT-5 minihigh· Epoch AI Benchmarking Hub data12.2%
92Kimi K2.7 Code· Epoch AI Benchmarking Hub data12.2%
93GPT-5.4 minixhigh· Epoch AI Benchmarking Hub data9.8%
94Claude Opus 4.532k· Epoch AI Benchmarking Hub data4.9%
95Inkling FP4xhigh· Epoch AI Benchmarking Hub data4.9%
96o4 Minihigh· Epoch AI Benchmarking Hub data4.9%
97Claude Opus 4.132k· Epoch AI Benchmarking Hub data2.4%
98Claude Sonnet 4.532k· Epoch AI Benchmarking Hub data2.4%
99DeepSeek V4 Promax· Epoch AI Benchmarking Hub data2.4%
100GPT-5 nanohigh· Epoch AI Benchmarking Hub data2.4%

Showing 51–100 of 104 results.

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Data by LLM Stats