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

50 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 Fable 5.1 (Max)Anthropic14.31%净改善率会话1513095% CI12.42% / 16.21%许可Proprietary会话1513095% CI12.42% / 16.21%许可Proprietary详情
02Claude Opus 5.5 (High)Anthropic13.82%净改善率会话535995% CI11.65% / 15.99%许可Proprietary会话535995% CI11.65% / 15.99%许可Proprietary来源
03Claude Sonnet 5.5 (Max)Anthropic12.52%净改善率会话521995% CI9.43% / 15.61%许可Proprietary会话521995% CI9.43% / 15.61%许可Proprietary来源
04GPT 6 Astra (Max)OpenAI12.27%净改善率会话1171795% CI10.04% / 14.50%许可Proprietary会话1171795% CI10.04% / 14.50%许可Proprietary详情
05GPT 6.1 Sol (Max)OpenAI11.23%净改善率会话627895% CI8.46% / 13.99%许可Proprietary会话627895% CI8.46% / 13.99%许可Proprietary来源
06GPT 6 Sol (Max)OpenAI9.71%净改善率会话610695% CI7.34% / 12.08%许可Proprietary会话610695% CI7.34% / 12.08%许可Proprietary来源
07Claude Opus 5 (High)Anthropic8.67%净改善率会话2756095% CI7.26% / 10.08%许可Proprietary会话2756095% CI7.26% / 10.08%许可Proprietary详情
08Claude Fable 5 (High)Anthropic8.21%净改善率会话4183295% CI7.00% / 9.43%许可Proprietary会话4183295% CI7.00% / 9.43%许可Proprietary详情
09Claude Opus 5 (Max)Anthropic7.92%净改善率会话2278195% CI6.45% / 9.40%许可Proprietary会话2278195% CI6.45% / 9.40%许可Proprietary详情
10Gemini 4 Argon (High)Google7.57%净改善率会话1003095% CI5.63% / 9.50%许可Proprietary会话1003095% CI5.63% / 9.50%许可Proprietary来源
11Claude Opus 4.8 (High)Anthropic6.64%净改善率会话4113295% CI5.30% / 7.98%许可Proprietary会话4113295% CI5.30% / 7.98%许可Proprietary详情
12GPT 5.6 Sol (xHigh)OpenAI6.45%净改善率会话3589695% CI5.17% / 7.73%许可Proprietary会话3589695% CI5.17% / 7.73%许可Proprietary详情
13Claude Sonnet 5 (High)Anthropic4.40%净改善率会话3246795% CI2.82% / 5.98%许可Proprietary会话3246795% CI2.82% / 5.98%许可Proprietary详情
14Kimi K3 (Max)Moonshot AI4.18%净改善率会话13250595% CI3.65% / 4.72%许可Kimi K3 license会话13250595% CI3.65% / 4.72%许可Kimi K3 license详情
15GPT 5.5 (xHigh)OpenAI4.15%净改善率会话5500695% CI3.13% / 5.18%许可Proprietary会话5500695% CI3.13% / 5.18%许可Proprietary详情
16Grok 4.7 (xHigh)xAI4.04%净改善率会话1159195% CI2.43% / 5.64%许可Proprietary会话1159195% CI2.43% / 5.64%许可Proprietary详情
17Deepseek V4.1 Flash (Max)DeepSeek4.02%净改善率会话11293895% CI3.56% / 4.49%许可MIT会话11293895% CI3.56% / 4.49%许可MIT详情
18Muse Spark 1.3 (Max)Meta4.00%净改善率会话6467195% CI3.40% / 4.59%许可Proprietary会话6467195% CI3.40% / 4.59%许可Proprietary详情
19Hy4 previewTencent3.95%净改善率会话5938895% CI3.16% / 4.75%许可Apache 2.0会话5938895% CI3.16% / 4.75%许可Apache 2.0详情
20GLM 5.2 (Max)Z.ai3.48%净改善率会话8617295% CI2.76% / 4.20%许可MIT会话8617295% CI2.76% / 4.20%许可MIT详情
21Gemini 3.8 Flash (High)Google2.97%净改善率会话3439695% CI2.12% / 3.81%许可Proprietary会话3439695% CI2.12% / 3.81%许可Proprietary详情
22Qwen3.8 MaxAlibaba2.51%净改善率会话5282495% CI1.89% / 3.14%许可Proprietary会话5282495% CI1.89% / 3.14%许可Proprietary详情
23GLM 5.3 (Max)Z.ai2.36%净改善率会话8638495% CI1.67% / 3.04%许可MIT会话8638495% CI1.67% / 3.04%许可MIT详情
24GPT 6 Luna (Max)OpenAI1.35%净改善率会话2924195% CI0.28% / 2.42%许可Proprietary会话2924195% CI0.28% / 2.42%许可Proprietary来源
25Grok 4.6 (xHigh)xAI1.28%净改善率会话2738095% CI0.28% / 2.28%许可Proprietary会话2738095% CI0.28% / 2.28%许可Proprietary详情
26Grok 4.5xAI1.21%净改善率会话4192095% CI0.20% / 2.22%许可Proprietary会话4192095% CI0.20% / 2.22%许可Proprietary详情
27DeepSeek V4 Pro (High) (0813)DeepSeek1.19%净改善率会话709095% CI-0.47% / 2.84%许可MIT会话709095% CI-0.47% / 2.84%许可MIT详情
28GPT 5.4 (High)OpenAI1.05%净改善率会话8159795% CI0.13% / 1.98%许可Proprietary会话8159795% CI0.13% / 1.98%许可Proprietary详情
29GPT 5.5OpenAI1.04%净改善率会话8241695% CI0.06% / 2.02%许可Proprietary会话8241695% CI0.06% / 2.02%许可Proprietary详情
30Step 5 PreviewStepFun0.48%净改善率会话945395% CI-0.99% / 1.95%许可Proprietary会话945395% CI-0.99% / 1.95%许可Proprietary来源
31GPT 5.6 Terra (xHigh)OpenAI0.40%净改善率会话2345795% CI-0.76% / 1.56%许可Proprietary会话2345795% CI-0.76% / 1.56%许可Proprietary详情
32GLM 5.3 FlashZ.ai-0.44%净改善率会话8083695% CI-0.90% / 0.03%许可MIT会话8083695% CI-0.90% / 0.03%许可MIT详情
33MiMo V2.6 FlashXiaomi-0.57%净改善率会话1527995% CI-1.85% / 0.72%许可MIT会话1527995% CI-1.85% / 0.72%许可MIT来源
34Qwen3.8 Flash NextAlibaba-0.73%净改善率会话8319895% CI-1.36% / -0.10%许可qwen-community-1.0会话8319895% CI-1.36% / -0.10%许可qwen-community-1.0详情
35GPT 5.6 Luna (xHigh)OpenAI-1.17%净改善率会话4259995% CI-1.84% / -0.49%许可Proprietary会话4259995% CI-1.84% / -0.49%许可Proprietary详情
36Gemini 3.7 Flash (High)Google-1.49%净改善率会话5459495% CI-2.22% / -0.75%许可Proprietary会话5459495% CI-2.22% / -0.75%许可Proprietary详情
37Qwen 3.8 27BAlibaba-1.66%净改善率会话5734395% CI-2.25% / -1.06%许可Apache 2.0会话5734395% CI-2.25% / -1.06%许可Apache 2.0详情
38Muse Spark 1.2 (xHigh)Meta-3.27%净改善率会话4815095% CI-3.95% / -2.60%许可Proprietary会话4815095% CI-3.95% / -2.60%许可Proprietary详情
39Muse Spark 1.1Meta-4.88%净改善率会话14271195% CI-5.33% / -4.44%许可Proprietary会话14271195% CI-5.33% / -4.44%许可Proprietary详情
40Qwen3.7 MaxAlibaba-5.13%净改善率会话4043895% CI-6.06% / -4.20%许可Proprietary会话4043895% CI-6.06% / -4.20%许可Proprietary详情
41Hy3Tencent-5.38%净改善率会话2967795% CI-6.59% / -4.17%许可Apache 2.0会话2967795% CI-6.59% / -4.17%许可Apache 2.0详情
42Minimax M3MiniMax-6.87%净改善率会话4126795% CI-7.76% / -5.99%许可MiniMax Community License会话4126795% CI-7.76% / -5.99%许可MiniMax Community License详情
43Qwen3.7 PlusAlibaba-7.15%净改善率会话2172295% CI-8.61% / -5.70%许可Proprietary会话2172295% CI-8.61% / -5.70%许可Proprietary详情
44Mimo V2.5 ProXiaomi-7.52%净改善率会话4523995% CI-8.41% / -6.63%许可MIT会话4523995% CI-8.41% / -6.63%许可MIT详情
45Gemini 3.6 Flash (High)Google-7.68%净改善率会话2291395% CI-8.74% / -6.63%许可Proprietary会话2291395% CI-8.74% / -6.63%许可Proprietary详情
46Gemini 3.1 Pro PreviewGoogle-7.71%净改善率会话8648395% CI-8.69% / -6.73%许可Proprietary会话8648395% CI-8.69% / -6.73%许可Proprietary详情
47Inkling Smallthinky-10.33%净改善率会话1327395% CI-11.73% / -8.92%许可Apache 2.0会话1327395% CI-11.73% / -8.92%许可Apache 2.0详情
48Inklingthinky-10.86%净改善率会话4281395% CI-11.90% / -9.82%许可Apache 2.0会话4281395% CI-11.90% / -9.82%许可Apache 2.0详情
49Mistral Medium 3.5Mistral AI-12.38%净改善率会话1115695% CI-13.90% / -10.86%许可Modified MIT会话1115695% CI-13.90% / -10.86%许可Modified MIT详情
50Solar Pro 4Upstage-15.90%净改善率会话950695% CI-17.75% / -14.05%许可Proprietary会话950695% CI-17.75% / -14.05%许可Proprietary详情

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 —