Skip to content

Benchmark

Arena ELO Image Edit - Multi Image Edit Agentic

Agentic benchmarks measure how well models carry out tasks that take several steps. Higher scores are better.

Models measured
19
Scores recorded
19
Sources
1
IQ weight
1.09

Score distribution

1,357.2
Mean
1,367.4
Median
103.9
Std dev
1,221.5
10th percentile
1,482.7
90th percentile
1,170.3
Min
1,170.31,549.9
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.09 for image-generation models, measured over 17 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT-Image-2.5 Sunburst· LMArena data1,549.9
2GPT Image 2Medium· LMArena data1,491.2
3GPT-Image-2.5 Flare· LMArena data1,482.7
4GPT-Image-2.6 FlareMedium· LMArena data1,470.6
5Seedream 5.0 Pro· LMArena data1,424.5
6Muse Image· LMArena data1,408.7
7Krea 3 Ultra· LMArena data1,402.7
8Gemini 3.1 Flash Imagenano-banana-2, Web Search· LMArena data1,391
9Gemini 3 Pro Imagenano-banana-pro, 2K· LMArena data1,389.7
10ChatGPT Image LatestHigh Fidelity, 20251216· LMArena data1,367.4
11GPT Image 1.5High Fidelity· LMArena data1,358.5
12MiniMax H3-Image· LMArena data1,351.8
13Ideogram V4.5· LMArena data1,319.7
14Doubao Seedream 4.5· LMArena data1,280.6
15UNI-1.1· LMArena data1,276.4
16Seedream 5.0 Lite· LMArena data1,248.5
17Gemini 2.5 Flash Imagenano-banana· LMArena data1,221.5
18FLUX.2 [pro]· LMArena data1,181.5
19Wan 2.6 Image· LMArena data1,170.3

Data credits: LMArena, CC BY 4.0 (licence)