M
ModelSage
排行榜/文档理解

AI文档理解排行榜

基于Arena Document Elo评分,找到文档解析和理解最强的大模型。

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

RankModelScore
1
Claude Opus 5 (high)
67.6
2
Claude Fable 5.1 (high with fallback)
67.2
3
Claude Opus 4.6 (max)
66.7
4
Claude Opus 4.6 (non-reasoning, high)
66.6
5
Claude Fable 5.1 (max with fallback)
65.4
6
Claude Opus 4.7 (max)
65.3
7
GPT-5.5 (xhigh)
64.3
8
GPT-5.5 (high)
64.2
9
GPT-5.6 Sol (xhigh)
64.0
10
Claude Sonnet 4.6 (non-reasoning, high)
63.9
11
GPT-5.4 (xhigh)
62.7
12
GPT-5.6 Terra (xhigh)
62.7
13
GPT-6 Astra (high)
62.3
14
Claude Sonnet 5 (high)
62.1
15
Claude Opus 4.8 (max)
61.9
16
Claude Opus 4.5
61.7
17
Gemini 3.5 Flash (medium)
61.6
18
GPT-5.6 Luna (xhigh)
61.1
19
Gemini 3.6 Flash (high)
61.0
20
Kimi K2.6
60.4
21
Qwen3.7 Plus
59.7
22
Gemini 3.1 Pro Preview
59.7
23
MiniMax-M3
58.7
24
Gemini 3 Pro Preview (high)
58.6
25
Kimi K2.5
58.2
26
Gemma 4 31BFREE
57.6
27
Gemini 2.5 Pro
57.2
28
GLM 5V Turbo
56.6
29
Gemini 3 Flash (non-reasoning)
56.3
30
GPT-5.5 Instant (June 2026)
55.2
31
GPT-5.1 (high)
55.2
32
GPT-5.2 (xhigh)
55.0