Skip to content

Benchmark

Arena ELO - Creative Writing

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

Models measured
364
Scores recorded
421
Sources
1
IQ weight
1.14

Score distribution

1,287.4
Mean
1,318.1
Median
144.2
Std dev
1,079.9
10th percentile
1,442.3
90th percentile
793.9
Min
793.91,532.7
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.14 for chat models, measured over 364 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
101Gemma 4 26B A4B· LMArena data1,401.8
102Qwen3 Max Preview· LMArena data1,401.7
103Gemini 3.1 Flash Lite· LMArena data1,401.6
104GLM 4.7· LMArena data1,401.2
105DeepSeek V3.1Thinking· LMArena data1,401.2
106GPT-6.1 TerraMax· LMArena data1,400.9
107Gemini 2.5 Flash· LMArena data1,400.2
108DeepSeek-V4-Flash PreviewHigh· LMArena data1,400.1
109Nemotron 3 UltraNVFP4· LMArena data1,399.9
110DeepSeek V3.2· LMArena data1,399.9
111GPT-5.2 Chat20260210· LMArena data1,399.1
112GPT-5.6 LunaxHigh· LMArena data1,396.5
113Grok 40709· LMArena data1,396.2
114GLM 4.5· LMArena data1,395.4
115Inkling FP4· LMArena data1,395
116Grok 4.1 FastReasoning· LMArena data1,395
117GPT-4.5 Preview2025-02-27· LMArena data1,393.8
118Hunyuan T120250711· LMArena data1,393
119MiMo V2 Omni· LMArena data1,392.9
120MiniMax M3· LMArena data1,391.4
121Mistral Medium 3.12508· LMArena data1,390.9
122MiMo V2.5· LMArena data1,390.7
123DeepSeek V3.1 TerminusThinking· LMArena data1,390.3
124Hunyuan Vision 1.5 Thinking· LMArena data1,389.5
125DeepSeek V3.2Thinking· LMArena data1,389
126DeepSeek V3.1· LMArena data1,387.9
127DeepSeek V3.2 ExpThinking· LMArena data1,387.9
128Beam 1.0 501B A23B· LMArena data1,387.5
129LongCat-Flash-Chat Exp2602· LMArena data1,387.4
130Claude Opus 4Thinking 16K, 20250514· LMArena data1,386.7
131Mistral Large 3· LMArena data1,386.3
132Grok 4 FastNon-Reasoning· LMArena data1,386.2
133Gemini 2.5 Flash Preview09-2025· LMArena data1,386.2
134Grok 4.7xHigh· LMArena data1,383.9
135Qwen 3 235B A22B2507, thinking· LMArena data1,383.6
136Kimi K2.5Instant· LMArena data1,382.3
137Qwen3 Max2025-09-23· LMArena data1,380.5
138Grok 4.3· LMArena data1,379.6
139NVIDIA Nemotron 3.5 Super VL 120BNVFP4· LMArena data1,379.5
140Qwen 3.8 27B· LMArena data1,378.1
141Grok 4 FastReasoning· LMArena data1,377.1
142MiMo V2 FlashNon-Thinking· LMArena data1,376.2
143GPT 5.2High· LMArena data1,375.5
144Qwen3 235B A22B Instruct 25072507· LMArena data1,374.9
145Claude Opus 420250514· LMArena data1,374.7
146GPT 5.2· LMArena data1,374.5
147Kimi K2 Thinking· LMArena data1,373.9
148Mistral Medium 3.5· LMArena data1,373.1
149GPT-6 SolMax· LMArena data1,371
150Claude Haiku 4.520251001· LMArena data1,370.2

Showing 101–150 of 421 results.

Data credits: LMArena, CC BY 4.0 (licence)