模型排行

下表是各机构榜单的真实排名,由本站定时抓取后直接显示在这里,无需跳转。分数由原机构按其方法计算,本站不重算、不混算;点击「数据来源」可查看完整榜单与口径。

综合能力

综合智能水平排名,分数由 Artificial Analysis 按其方法计算。
数据来源:Artificial Analysis ↗ · 更新于 2026-11-18 01:40

#模型机构 / 类别智能指数备注
01Claude Fable 5.2 Max, Default FallbackAnthropicAnthropic65$26.54
02Gemini 4 Carbon HighGoogleGoogle60$5.48
03Claude Opus 5.5 Max, Default FallbackAnthropicAnthropic59$10.04
04GPT-6 Astra MaxOpenAIOpenAI57$6.24
05GPT-6.1 Sol MaxOpenAIOpenAI55$1.27
06Claude Sonnet 5.5 Max, Default FallbackAnthropicAnthropic55$10.02
07Gemini 4 Hydrogen XhighGoogleGoogle54$3.16
08GPT-6.1 Terra MaxOpenAIOpenAI51$0.96
09Gemini 4 Argon HighGoogleGoogle50$4.97
10Muse Wonder Full Max50--

编程能力

SWE-bench Verified 上真实 GitHub issue 的解决比例。
数据来源:SWE-bench Verified ↗ · 更新于 2026-11-18 01:40

#模型机构 / 类别解决率备注
01Sonar Foundation Agent + Claude 4.5 OpusAnthropicAnthropic79.2%
02live-SWE-agent + Claude 4.5 Opus medium (20251101)AnthropicAnthropic79.2%
03TRAE + Doubao-Seed-CodeByteDanceByteDance78.8%
04live-SWE-agent + Gemini 3 Pro Preview (2025-11-18)Google DeepMindGoogle DeepMind77.4%
05EPAM AI/Run Developer Agent v20250719 + Claude 4 SonnetAnthropicAnthropic76.8%
06Atlassian Rovo Dev (2025-09-02)AnthropicAnthropic76.8%
07ACoderAnthropicAnthropic76.4%
08WarpOpenAIOpenAI75.6%
09TRAE + Claude Sonnet 4 + Opus 4 + Sonnet 3.7 + Gemini 2.5 ProAnthropicAnthropic75.2%
10Harness AIAnthropicAnthropic74.8%

画图能力

文生图模型的盲测偏好排名:同题两图由人投票,Elo 由胜负累积而来。
数据来源:Artificial Analysis ↗ · 更新于 2026-11-18 01:40

#模型机构 / 类别Elo 评分备注
01GPT Image 2.6 Sunburst (max)OpenAIOpenAI1261$210.72 /1k imgs
02GPT Image 2.5 Sunburst (max)OpenAIOpenAI1229$210.72 /1k imgs
03GPT Image 2.6 Flare (max)OpenAIOpenAI1227$210.72 /1k imgs
04GPT Image 2.5 Flare (max)OpenAIOpenAI1210$210.72 /1k imgs
05GPT Image 2 (high)OpenAIOpenAI1192$211.00 /1k imgs
06Kling Image 4.0KlingAIKlingAI1180Coming soon
07Nano Banana 2.1GoogleGoogle1176$33.60 /1k imgs
08MAI-Image-2.6Microsoft AIMicrosoft AI1168$38.91 /1k imgs
09Grok Imagine Image 2.0SpaceXAISpaceXAI1167$60.00 /1k imgs
10Canva CraftCanvaCanva1158$30.00 /1k imgs

视频能力

文生视频模型的盲测偏好排名:同题两段视频由人投票,Elo 由胜负累积而来。
数据来源:Artificial Analysis ↗ · 更新于 2026-11-18 01:40

#模型机构 / 类别Elo 评分备注
01Wan 3.0AlibabaAlibaba1174$12.00 /min
02Utopai XUtopai StudiosUtopai Studios1168No API available
03MiniMax H3 (768p)MiniMaxMiniMax1159$4.80 /min
04Dreamina Seedance 2.5ByteDance SeedByteDance Seed1156$34.12 /min
05Hyperflow H3Video RebirthVideo Rebirth1154Coming soon
06FLUX 3 MaxFalFal1152$8.70 /min
07Gemini Omni Flash 1.2GoogleGoogle1151$6.26 /min
08HeyGen Video 1HeyGenHeyGen1149$2.40 /min
09MiniMax H3 MaxFalFal1143$9.60 /min
10FLUX 3Black Forest LabsBlack Forest Labs1136$17.40 /min

检索与嵌入

嵌入 / 检索模型的社区下载量,反映采用热度。
数据来源:Hugging Face ↗ · 更新于 2026-11-18 01:40

#模型机构 / 类别月下载量备注
01bge-small-en-v1.5BAAIBAAI44.3MEmbedding
02bge-base-en-v1.5BAAIBAAI8.6MEmbedding
03Qwen3-Embedding-0.6BQwenQwen8.0MEmbedding
04multilingual-e5-largeintfloatintfloat7.3MEmbedding
05bge-large-en-v1.5BAAIBAAI7.2MEmbedding
06granite-embedding-small-english-r2ibm-graniteibm-granite6.7MEmbedding
07all-MiniLM-L6-v2XenovaXenova4.6MEmbedding
08bge-base-en-v1.5XenovaXenova2.6MEmbedding
09bge-small-zh-v1.5BAAIBAAI2.6MEmbedding
10Qwen3-Embedding-4BQwenQwen2.3MEmbedding

开源与实际使用

OpenRouter 上真实 API 请求处理的 token 量,反映开发者实际使用。
数据来源:OpenRouter Rankings ↗ · 更新于 2026-11-18 01:40

#模型机构 / 类别处理 token 数备注
01DeepSeek V4.1 Flash 1017deepseekdeepseek61.2T tokens+10%
02Hy4tencenttencent16T tokens+0%
03DeepSeek V4.1 Flashdeepseekdeepseek15.5T tokens-12%
04GLM 5.3 Flashz-aiz-ai11.8T tokens-3%
05GPT-6 Lunaopenaiopenai8.08T tokens+14%
06Claude Opus 5.5anthropicanthropic7.37T tokens+9%
07Step 5 Previewstepfunstepfun6.36T tokens-11%
08MiMo-V2.6-Flashxiaomixiaomi4.16T tokens+63%
09GLM 5.3z-aiz-ai3.74T tokens+5%
10GPT-6.1 Solopenaiopenai3.71T tokens-5%