M
ModelSage
排行榜/视觉/多模态

AI视觉能力排行榜

基于Arena Vision Elo评分排名，找到视觉理解最强的大模型。

分数怎么算：分数 = (Elo − 900) ÷ (全站最高 Elo − 900) × 100，与其他 Arena 榜共用一把尺子。

Arena AI 数据更新于 2026-10-14，每天北京时间凌晨自动刷新

综合能力代码能力Agent 智能体Arena 人类偏好视觉/多模态搜索能力文档理解响应速度性价比
RankModelScore
1
Claude Fable 5.1 (max with fallback)
44.8
2
Qwen3.8 Max (0902)
44.0
3
Claude Opus 4.6 (max)
43.8
4
Claude Opus 4.7 (max)
43.7
5
Gemini 3.7 Flash (medium)
43.6
6
Claude Opus 4.6 (non-reasoning, high)
43.3
7
GPT-6.1 Sol (high)
43.3
8
Claude Opus 5.5 (high with fallback)
43.1
9
Gemini 3.8 Flash (high)
42.7
10
Gemini 3 Pro Preview (high)
42.7
11
Claude Opus 5 (high)
42.5
12
GPT-5.6 Sol (xhigh)
42.5
13
Claude Fable 5.1 (high with fallback)
42.4
14
GPT-6 Astra (high)
42.4
15
GPT-5.5 (xhigh)
42.3
16
Gemini 3.5 Flash (medium)
42.1
17
Gemini 3.6 Flash (high)
41.8
18
GPT-5.5 (high)
41.8
19
Gemini 3.1 Pro Preview
41.7
20
GPT-5.4 (xhigh)
41.6
21
Claude Sonnet 4.6 (non-reasoning, high)
41.2
22
GPT-5.5 Instant (June 2026)
41.2
23
GLM-5.3-Flash
41.0
24
Gemini 3 Flash (non-reasoning)
40.9
25
GPT-6 Sol (high)
40.7
26
GPT-5.6 Terra (xhigh)
40.4
27
Claude Sonnet 5.5 (xhigh with fallback)
40.2
28
Gemini 3.5 Flash-Lite
40.2
29
Kimi K2.6
40.1
30
Qwen3.7 Plus
40.0
31
Claude Sonnet 5 (high)
39.9
32
Gemma 4 31BFREE
39.7
33
DeepSeek V4.1 Flash (max)
39.5
34
Gemini 3 Flash
39.5
35
GPT-5.6 Luna (xhigh)
39.3
36
Kimi K2.5
38.6
37
GPT-5.4 mini (xhigh)
38.5
38
Gemini 2.5 Pro
38.2
39
MiMo-V2.6-Flash
38.2
40
Qwen3.5 397B A17B
38.0
41
MiMo-V2.6-Pro
38.0
42
Qwen3.8 27B (xhigh)
37.8
43
Gemma 4 26B A4B
37.5
44
GPT-6 Luna (high)
37.4
45
GPT-5.1 (high)
37.3
46
GPT-4o (Nov)
37.2
47
Kimi K2 Thinking
36.9
48
MiniMax-M3
36.8
49
Gemini 3.1 Flash-Lite
36.7
50
MiMo-V2.5-Pro
36.6
51
GLM 5V Turbo
36.3
52
GPT-5.2 (xhigh)
36.1
53
Qwen3.5 122B A10B
35.8
54
GPT-5 (low)
35.8
55
GPT-4
35.7
56
Qwen3.5 27B
34.8
57
Qwen3 VL 235B A22B (指令)
34.6
58
Gemini 2.5 Flash (non-reasoning)
34.6
59
o3
34.6
60
GPT-4.1
34.5
61
GPT-5.6 Sol (max)
34.0
62
o4-mini (high)
33.0
63
GPT-4.1 mini
33.0
64
GPT-5.4 nano (xhigh)
32.9
65
o1-preview
32.1
66
Qwen3 VL 235B A22B (推理)
31.7
67
Claude Opus 4.8 (max)
31.7
68
Gemini 2.5 Flash-Lite (non-reasoning)
31.7
69
Claude Sonnet 4.6 (max)
31.6
70
GPT-5 mini (medium)
31.1
71
Claude 3.7 Sonnet (non-reasoning)
30.2
72
Gemini 2.5 Flash-Lite (Sep) (non-reasoning)
30.1
73
GPT-4o (May)
28.7
74
GLM-4.6V (non-reasoning)
28.6
75
Gemma 3 27BFREE
28.5
76
GLM-4.5V (non-reasoning)
27.8
77
GPT-5 nano (high)
27.0
78
Claude 3.5 Sonnet (Oct)
26.9
79
GPT-4o (Aug)
23.9
80
GPT-4 Turbo
23.2
81
GPT-4o mini
21.6
82
GPT-4.1 nano
20.7
83
Claude 3 Opus
17.7
84
Claude 3 Haiku
10.9