AI 智能体模型排行榜

查看 Arena 智能体任务榜的净改善率。

评测来源 Arena · 智能体原榜发布 查看原榜评分方法

净改善率

58 个模型
Arena · 智能体 · 净改善率
排名模型 / 开发机构净改善率会话95% CI许可详情
01Claude Opus 5.5 (Max)Anthropic20.14%净改善率会话1416995% CI18.06% / 22.23%许可Proprietary会话1416995% CI18.06% / 22.23%许可Proprietary来源
02Gemini 4 Carbon (High)Google16.72%净改善率会话4487395% CI15.95% / 17.48%许可Proprietary会话4487395% CI15.95% / 17.48%许可Proprietary来源
03Claude Sonnet 5.5 (Max)Anthropic15.42%净改善率会话859695% CI12.65% / 18.20%许可Proprietary会话859695% CI12.65% / 18.20%许可Proprietary来源
04GPT-6 Astra (Max)OpenAI14.70%净改善率会话1547595% CI12.63% / 16.76%许可Proprietary会话1547595% CI12.63% / 16.76%许可Proprietary来源
05Claude Fable 5.1 (Max)Anthropic13.86%净改善率会话1751895% CI11.98% / 15.73%许可Proprietary会话1751895% CI11.98% / 15.73%许可Proprietary详情
06GPT-6.1 Sol (Max)OpenAI12.51%净改善率会话3028595% CI11.41% / 13.62%许可Proprietary会话3028595% CI11.41% / 13.62%许可Proprietary来源
07GPT-6.1 Terra (Max)OpenAI11.40%净改善率会话2135595% CI10.00% / 12.79%许可Proprietary会话2135595% CI10.00% / 12.79%许可Proprietary来源
08Gemini 4 Argon (High)Google10.53%净改善率会话5281895% CI9.68% / 11.37%许可Proprietary会话5281895% CI9.68% / 11.37%许可Proprietary来源
09Muse Wonder Full (Max)Meta9.42%净改善率会话3259295% CI8.48% / 10.37%许可Proprietary会话3259295% CI8.48% / 10.37%许可Proprietary来源
10Claude Fable 5 (High)Anthropic7.82%净改善率会话5125195% CI6.71% / 8.94%许可Proprietary会话5125195% CI6.71% / 8.94%许可Proprietary详情
11Claude Opus 5 (Max)Anthropic7.66%净改善率会话3204895% CI6.42% / 8.91%许可Proprietary会话3204895% CI6.42% / 8.91%许可Proprietary详情
12GPT-6 Sol (Max)OpenAI6.78%净改善率会话1267095% CI5.23% / 8.32%许可Proprietary会话1267095% CI5.23% / 8.32%许可Proprietary来源
13Hy4Tencent5.39%净改善率会话1270095% CI3.68% / 7.10%许可Proprietary会话1270095% CI3.68% / 7.10%许可Proprietary来源
14Kimi K3.1 Pro (Max)Moonshot AI5.29%净改善率会话1804595% CI4.01% / 6.57%许可Kimi K3.1 Pro License会话1804595% CI4.01% / 6.57%许可Kimi K3.1 Pro License来源
15GPT-5.6 Sol (xHigh)OpenAI4.57%净改善率会话4553395% CI3.54% / 5.61%许可Proprietary会话4553395% CI3.54% / 5.61%许可Proprietary来源
16MiMo-V2.6-ProXiaomi2.68%净改善率会话4873595% CI1.77% / 3.60%许可MIT会话4873595% CI1.77% / 3.60%许可MIT来源
17Grok 4.7 (xHigh)xAI2.68%净改善率会话1891795% CI1.22% / 4.14%许可Proprietary会话1891795% CI1.22% / 4.14%许可Proprietary详情
18Gemini 3.8 Flash (High)Google2.58%净改善率会话5306895% CI1.75% / 3.41%许可Proprietary会话5306895% CI1.75% / 3.41%许可Proprietary详情
19Muse Spark 1.3 (Max)Meta2.51%净改善率会话8105395% CI1.68% / 3.34%许可Proprietary会话8105395% CI1.68% / 3.34%许可Proprietary详情
20Claude Sonnet 5 (High)Anthropic2.44%净改善率会话4185795% CI1.28% / 3.61%许可Proprietary会话4185795% CI1.28% / 3.61%许可Proprietary详情
21Muse Wonder (Max)Meta2.43%净改善率会话1336895% CI1.08% / 3.78%许可Proprietary会话1336895% CI1.08% / 3.78%许可Proprietary来源
22DeepSeek-V4.1-Flash-20261017 (Max)DeepSeek2.37%净改善率会话7088795% CI1.80% / 2.94%许可MIT会话7088795% CI1.80% / 2.94%许可MIT来源
23GPT-6 Luna (Max)OpenAI1.89%净改善率会话22652895% CI1.45% / 2.33%许可Proprietary会话22652895% CI1.45% / 2.33%许可Proprietary来源
24GLM-5.3 (Max)Z.ai1.09%净改善率会话9628295% CI0.29% / 1.88%许可MIT会话9628295% CI0.29% / 1.88%许可MIT来源
25Kimi K3 (Max)Moonshot AI0.84%净改善率会话16133395% CI0.26% / 1.42%许可Kimi K3 license会话16133395% CI0.26% / 1.42%许可Kimi K3 license详情
26Hy4 previewTencent0.75%净改善率会话10147895% CI0.05% / 1.44%许可Apache 2.0会话10147895% CI0.05% / 1.44%许可Apache 2.0详情
27GLM-5.2 (Max)Z.ai0.64%净改善率会话9931595% CI-0.13% / 1.41%许可MIT会话9931595% CI-0.13% / 1.41%许可MIT来源
28Claude Haiku 5.5 (High)Anthropic0.62%净改善率会话3120495% CI-0.40% / 1.65%许可Proprietary会话3120495% CI-0.40% / 1.65%许可Proprietary来源
29Qwen3.8-MaxAlibaba0.48%净改善率会话8021095% CI-0.18% / 1.13%许可Proprietary会话8021095% CI-0.18% / 1.13%许可Proprietary来源
30Grok 4.6 (xHigh)xAI0.45%净改善率会话3039895% CI-0.72% / 1.63%许可Proprietary会话3039895% CI-0.72% / 1.63%许可Proprietary详情
31MiMo-V2.6-FlashXiaomi0.20%净改善率会话3094695% CI-0.93% / 1.33%许可MIT会话3094695% CI-0.93% / 1.33%许可MIT来源
32DeepSeek-V4-Pro (High, 20260813)DeepSeek-0.30%净改善率会话770895% CI-2.00% / 1.40%许可MIT会话770895% CI-2.00% / 1.40%许可MIT来源
33GPT-5.6 Terra (xHigh)OpenAI-1.50%净改善率会话4468395% CI-2.29% / -0.72%许可Proprietary会话4468395% CI-2.29% / -0.72%许可Proprietary来源
34Gemini 3.7 Flash (High)Google-1.60%净改善率会话6373995% CI-2.48% / -0.73%许可Proprietary会话6373995% CI-2.48% / -0.73%许可Proprietary详情
35GPT-6 Luna (Medium)OpenAI-1.84%净改善率会话6715095% CI-2.37% / -1.31%许可Proprietary会话6715095% CI-2.37% / -1.31%许可Proprietary来源
36GLM-5.3-FlashZ.ai-2.39%净改善率会话9592495% CI-3.09% / -1.70%许可MIT会话9592495% CI-3.09% / -1.70%许可MIT来源
37MiniMax-M3.1-FlashMiniMax-2.84%净改善率会话1251695% CI-4.38% / -1.30%许可MiniMax Community License会话1251695% CI-4.38% / -1.30%许可MiniMax Community License来源
38Step 5 PreviewStepFun-2.88%净改善率会话1699295% CI-4.27% / -1.48%许可Proprietary会话1699295% CI-4.27% / -1.48%许可Proprietary来源
39Qwen3.8-27BAlibaba-4.28%净改善率会话6428595% CI-5.13% / -3.42%许可Apache 2.0会话6428595% CI-5.13% / -3.42%许可Apache 2.0来源
40Qwen3.8-Flash-NextAlibaba-4.43%净改善率会话11002395% CI-5.16% / -3.70%许可qwen-community-1.0会话11002395% CI-5.16% / -3.70%许可qwen-community-1.0来源
41Gemini 3.1 Pro PreviewGoogle-4.44%净改善率会话8806095% CI-5.86% / -3.03%许可Proprietary会话8806095% CI-5.86% / -3.03%许可Proprietary详情
42Beam 1.0 501B A23Breflection-5.38%净改善率会话1242295% CI-7.89% / -2.88%许可Apache 2.0会话1242295% CI-7.89% / -2.88%许可Apache 2.0来源
43Mistral Large 4Mistral AI-5.46%净改善率会话549995% CI-7.39% / -3.53%许可Modified MIT会话549995% CI-7.39% / -3.53%许可Modified MIT来源
44GPT-6 Luna (Low)OpenAI-5.93%净改善率会话4181295% CI-6.63% / -5.22%许可Proprietary会话4181295% CI-6.63% / -5.22%许可Proprietary来源
45Qwen3.7-MaxAlibaba-6.52%净改善率会话4330795% CI-7.73% / -5.31%许可Proprietary会话4330795% CI-7.73% / -5.31%许可Proprietary来源
46Qwen3.7-PlusAlibaba-7.46%净改善率会话2457795% CI-9.02% / -5.90%许可Proprietary会话2457795% CI-9.02% / -5.90%许可Proprietary来源
47Muse Spark 1.1Meta-7.66%净改善率会话15050595% CI-8.32% / -7.00%许可Proprietary会话15050595% CI-8.32% / -7.00%许可Proprietary详情
48Muse Spark 1.2 (xHigh)Meta-7.70%净改善率会话5509095% CI-8.54% / -6.85%许可Proprietary会话5509095% CI-8.54% / -6.85%许可Proprietary详情
49Hy3Tencent-7.72%净改善率会话3277395% CI-9.17% / -6.27%许可Apache 2.0会话3277395% CI-9.17% / -6.27%许可Apache 2.0详情
50MiMo-V2.5-ProXiaomi-9.03%净改善率会话4964495% CI-10.04% / -8.01%许可MIT会话4964495% CI-10.04% / -8.01%许可MIT来源
显示更多已显示 50 / 58

数据来自 Arena 官方公开数据集(CC BY 4.0);保留原榜名次与评分。 官方数据集

未公布的数据记为 —