Skip to content

Benchmark

Arena ELO - Industry Mathematical

People compare model answers and choose which they prefer. Higher scores are better.

Models measured
348
Scores recorded
402
Sources
1
IQ weight
1.25

Score distribution

1,321.4
Mean
1,367.7
Median
154.1
Std dev
1,094.5
10th percentile
1,481.4
90th percentile
833.6
Min
833.61,566.4
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.25 for chat models, measured over 348 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51Gemma 4 26B A4B· LMArena data1,474.5
52Muse Spark 1.2xHigh· LMArena data1,473.7
53Kimi K2.5Thinking· LMArena data1,472.6
54Gemini 3 Pro Preview· LMArena data1,471.9
55Qwen3.6 Max Preview· LMArena data1,471.6
56GLM 5.1· LMArena data1,470.8
57Qwen3.7 Plus· LMArena data1,469.2
58Claude Opus 4.520251101· LMArena data1,468.4
59GPT-5.6 LunaxHigh· LMArena data1,467.7
60GPT 5.4· LMArena data1,466.9
61Inkling FP4· LMArena data1,466.2
62Beam 1.0 501B A23B· LMArena data1,466.2
63Step 5 Preview· LMArena data1,465.7
64MiMo V2 Pro· LMArena data1,464.3
65Gemini 3 Flash· LMArena data1,464.2
66Gemma 4 31B IT· LMArena data1,464.1
67Grok 4.6High· LMArena data1,464.1
68Claude Sonnet 4.5High, 32K, 20250929· LMArena data1,462.6
69Qwen3.7 Max Preview· LMArena data1,462.1
70Qwen 3.8 27B· LMArena data1,461.6
71Nemotron 3 UltraNVFP4· LMArena data1,461.3
72Claude Opus 4.5High, 32K, 20251101· LMArena data1,458.5
73Mistral Large 4· LMArena data1,456.3
74Gemini 2.5 Pro· LMArena data1,455.4
75DeepSeek V4 Pro· LMArena data1,455.2
76Muse Spark· LMArena data1,454.7
77DeepSeek-V4-Pro PreviewHigh· LMArena data1,453.9
78Qwen3.5 397B A17B· LMArena data1,452.7
79MiMo V2 Omni· LMArena data1,450.4
80GPT 5.1High· LMArena data1,450
81MiMo V2.5· LMArena data1,449.1
82GLM 5· LMArena data1,448.8
83Qwen3.6 Plus· LMArena data1,448.1
84GLM 4.6· LMArena data1,447.8
85Kimi K3.1 ProMax· LMArena data1,447.6
86Grok 4.20Reasoning, 0309, beta· LMArena data1,446.3
87Claude Sonnet 4.520250929· LMArena data1,445.3
88Qwen3 Max Preview· LMArena data1,444.8
89Grok 4.20 Multi-Agent0309, beta· LMArena data1,442.4
90GPT 5.2High· LMArena data1,442
91Qwen3 235B A22B Instruct 25072507· LMArena data1,441.2
92MiniMax M3· LMArena data1,441
93nova-experimental-chat26-01-10· LMArena data1,441
94LongCat Flash Chat· LMArena data1,440.5
95Mistral Medium 3.5· LMArena data1,440.2
96DeepSeek V3.2· LMArena data1,438.3
97DeepSeek V3.1Thinking· LMArena data1,438.1
98DeepSeek V3.2Thinking· LMArena data1,438.1
99MiniMax M2.7· LMArena data1,438.1
100Kimi K2.5Instant· LMArena data1,437.7

Showing 51–100 of 402 results.

Data credits: LMArena, CC BY 4.0 (licence)