1
O
GPT 5.5 (High)
GPT 5.5 (High)Openai
100.0
24.6K
1.05M
¥36 / ¥216Input/Output
Agent · Overall · Overall Leaderboard
Ranking for Overall / Overall, based on public preference data.
Overall model ranking guide
Ranking for Overall / Overall, based on public preference data.
GPT 5.5 (High)Claude Opus 4.7 (Thinking)GPT 5.4 (High)Claude Opus 4.6GPT 5.5
2
A
Claude Opus 4.7 (Thinking)
Claude Opus 4.7 (Thinking)Anthropic
94.1
24.5K
1M
¥36 / ¥180Input/Output
3
OGPT 5.4 (High)
GPT 5.4 (High)Openai
88.2
24.4K
1.05M
¥18 / ¥108Input/Output
4
AClaude Opus 4.6
Claude Opus 4.6Anthropic
82.4
24.7K
1M
¥36 / ¥180Input/Output
5
OGPT 5.5
GPT 5.5Openai
76.5
24.9K
1.05M
¥36 / ¥216Input/Output
6
AClaude Opus 4.7
Claude Opus 4.7Anthropic
70.6
24.7K
1M
¥36 / ¥180Input/Output
7
AClaude Sonnet 4.6
Claude Sonnet 4.6Anthropic
64.7
24.6K
1M
¥21.6 / ¥108Input/Output
8
Z
GLM 5.1
GLM 5.1Zai
58.8
19.8K
200K
¥0 / ¥0Input/Output
9
G
Gemini 3.1 Pro Preview
Gemini 3.1 Pro PreviewGoogle
52.9
24.5K
1.05M
¥14.4 / ¥86.4Input/Output
10
GGemini 3.5 Flash
Gemini 3.5 FlashGoogle
47.1
17.7K
1.05M
¥10.8 / ¥64.8Input/Output
11
M
Kimi K2.6
Kimi K2.6Moonshot
41.2
21.3K
262K
¥6.84 / ¥28.8Input/Output
12
D
DeepSeek V4 Pro
DeepSeek V4 ProDeepseek
35.3
20K
1M
¥3.13 / ¥6.26Input/Output
13
A
Qwen 3.6 Plus
Qwen 3.6 PlusAlibaba
29.4
19.5K
1M
¥3.6 / ¥21.6Input/Output
14
DDeepSeek V4 Flash
DeepSeek V4 FlashDeepseek
23.5
19.9K
1M
¥1.01 / ¥2.02Input/Output
15
M
Minimax M2.7
Minimax M2.7Minimax
17.6
20K
205K
¥0 / ¥0Input/Output
16
GGemini 3 Flash
Gemini 3 FlashGoogle
11.8
24.5K
1.05M
¥3.6 / ¥21.6Input/Output
17
GGemma 4 31B
Gemma 4 31BGoogle
5.9
13.7K
262K
¥3.24 / ¥7.2Input/Output
18
X
Grok 4.3
Grok 4.3Xai
0.0
23.7K
1M
¥9 / ¥18Input/Output
GPT 5.5 (High) why it ranks first
GPT 5.5 (High) ranks first with a percent score of 100.0 and 24.6K samples. Use it as the first option for this leaderboard, then compare price, context and availability.
Do not only look at rank #1
Start with the leaderboard closest to your task. Compare the top models by score and sample size, then check price, context length, open or closed access, and provider availability.
Compare adjacent capabilities
FAQ
智能体总榜排行榜看什么指标？
主要看排名、百分制分数、样本量和来源。分数用于快速比较同一榜单内模型表现，样本量用于判断结果稳定性。
为什么不同榜单不能直接混合成总分？
不同榜单的任务、样本和评测口径不同，模力榜默认只在同一榜单内排序，避免把写作、代码、图像等能力强行合并。
智能体总榜模型应该怎么选？
优先看与你任务最接近的榜单，再结合价格、上下文长度、开源闭源和厂商可用性。排名靠前不代表适合所有预算和部署方式。
榜单多久更新？
页面展示的是最新成功采集的公开榜单数据。当前优先使用 LMArena leaderboard dataset，并在页面来源中保留原始链接。