BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

Arena Vision

Pairwise votes on image-understanding prompts.

As of 24 Oct 2026, Claude Fable 5 leads Arena Vision on BenchLeader with 1325, ahead of Claude Opus 5.5 at 1321, across 159 model configurations with a published result.

Published by
Arena (LMArena)
Category
Multimodal
Index weight
1.0
Models
159
Data as of
24 Oct 2026
Newest result
19 Oct 2026

CC BY 4.0 — lmarena-ai/leaderboard-dataset on Hugging Face.

What the test looks like

Pairwise votes on prompts that include an image: describing, reading, or reasoning about pictures and screenshots.

How it is scored

Bradley-Terry rating on image prompts.

What to keep in mind

The prompt mix is whatever visitors upload, skewed to screenshots and memes rather than medical or technical imagery. Only models with vision are listed.

Top 15
  1. 1Claude Fable 5 (high)1325
  2. 2Claude Opus 5.5 (high)1321
  3. 3Claude Opus 5 (high)1320
  4. 4Claude Fable 5.1 (max)1320
  5. 5Claude Opus 4.71317
  6. 6Claude Opus 4.6 (high)1316
  7. 7Gemini 3.8 Flash (high)1315
  8. 8Gemini 3.7 Flash (high)1314
  9. 9Claude Opus 4.7 (high)1313
  10. 10Qwen3.8 Max (max)1313
  11. 11Claude Opus 4.61312
  12. 12Muse Spark 1.3 (max)1308
  13. 13Gemini 3.5 Flash (high)1308
  14. 14Gemini 3.5 Flash (medium)1306
  15. 15Muse Spark1306
159 of 159
#
1Claude Fable 5high132564.62026-10-19
2Claude Opus 5.5high132169.42026-10-19
3Claude Opus 5high132067.92026-10-19
4Claude Fable 5.1max132070.52026-10-19
5Claude Opus 4.7131763.32026-10-19
6Claude Opus 4.6high131662.72026-10-19
7Gemini 3.8 Flashhigh131564.42026-10-19
8Gemini 3.7 Flashhigh131463.52026-10-19
9Claude Opus 4.7high131361.02026-10-19
10Qwen3.8 Maxmax131363.32026-10-19
11Claude Opus 4.6131262.62026-10-19
12Muse Spark 1.3max130867.12026-10-19
13Gemini 3.5 Flashhigh130863.12026-10-19
14Gemini 3.5 Flashmedium130662.82026-10-19
15Muse Spark130664.32026-10-19
16Gemini 3 Pro130560.42026-10-19
17Muse Spark 1.2xhigh130463.22026-10-19
18GPT-5.4high130357.92026-10-19
19Muse Spark 1.3xhigh130264.92026-10-19
20Gemini 3.6 Flashhigh129962.02026-10-19
21GPT-5.5129760.32026-10-19
22Gemini 3.1 Pro129660.62026-10-19
23Claude Opus 4.8high129560.72026-10-19
24GLM 5.3 Flash129462.62026-10-19
25GPT-5.5high129365.32026-10-19
26Muse Spark 1.1129362.52026-10-19
27GPT-5.4129258.22026-10-19
28Claude Opus 4.8128960.12026-10-19
29GPT-6.1 Solmax128869.92026-10-19
30Claude Sonnet 5.5xhigh128767.92026-10-19
31Grok 4.5128759.02026-10-19
32Gemini 3 Flash128557.02026-10-19
33Claude Sonnet 4.6128357.12026-10-19
34Kimi K2.6128255.42026-10-19
35GPT-5.6 Solxhigh128166.62026-10-19
36GPT-6 Astramax128070.42026-10-19
37Qwen3.7 Plus127858.52026-10-19
38DeepSeek V4.1 Flashmax127861.22026-10-19
39Gemma 4 31B127748.42026-10-19
40Dola Seed 2.0 Pro127458.52026-10-19
41Claude Sonnet 5high127360.22026-10-19
42Mistral Large 4127256.62026-10-19
43GPT-5.6 Terraxhigh127162.82026-10-19
44Qwen3.8 27B127054.62026-10-19
45Gemini 3.5 Flash Lite126954.52026-10-19
46Kimi K2.5thinking126957.62026-10-19
47GPT-5.2126858.02026-10-19
48Gemini 3 Flashminimal126654.02026-10-19
49Step 5 Preview126560.42026-10-19
50GLM 5V Turbo126455.92026-10-19
51Gemini 2.5 Pro126353.92026-10-19
52Grok 4.6high126362.92026-10-19
53MiMo-V2.6-Pro126364.02026-10-19
54Qwen3.5 397B A17B126351.52026-10-19
55Grok 4.20thinking126258.62026-10-19
56Gemma 4 26B A4B126056.22026-10-19
57Grok 4.20 Multi-Agent125958.52026-10-19
58GPT-5.6 Lunaxhigh125759.42026-10-19
59MiMo-V2.6-Flash125659.72026-10-19
60Kimi K2.5125454.52026-10-19

More multimodal benchmarks · All LLM benchmarks

Cite as: BenchLeader, “Arena Vision leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/lmarena_vision.

Arena Vision: questions

What does Arena Vision measure?
Pairwise votes on prompts that include an image: describing, reading, or reasoning about pictures and screenshots. Scores are reported in Elo-style ratings from pairwise comparisons; higher is better.
Which AI model leads Arena Vision?
Claude Fable 5 leads Arena Vision with 1325 as of 24 Oct 2026, ahead of Claude Opus 5.5 at 1321.
How many models have Arena Vision results?
159 model configurations have a Arena Vision result on BenchLeader, all taken from Arena (LMArena).
Who runs Arena Vision and how often is it updated?
Arena Vision is published by Arena (LMArena). BenchLeader re-reads the published results once a day and records the date each result was published.
Does Arena Vision count toward the BenchLeader Index?
Yes. Arena Vision contributes to the multimodal category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.