88Z.ai / 排行榜

AI 模型速度与延迟排行榜

采用 Artificial Analysis 公布的每任务耗时和首 token 延迟,耗时越低越好。

本站最近采集时间:2026-11-14 06:41 (UTC+8, Asia/Shanghai)。各来源的发布日期见表中或文末引用。

采用 Artificial Analysis 公布的每任务耗时和首 token 延迟,耗时越低越好。

55 个模型系列(同一系列只展示成绩最好的变体)

表头可点击排序。以下为服务器渲染的完整表格,关闭脚本后仍可阅读。

#模型每任务耗时(分钟)首 token 延迟(秒)AA Index
1GPT 6 Luna (Low)OpenAI0.63.3410
2GPT 6.1 Terra (Low)OpenAI1.22.8329
3Gemini 4 Hydrogen (Low)Google1.35.5627
4Mistral Small 4Mistral1.69.954
5Gemini 3.5 Flash LiteGoogle1.74.6112
6GPT 6.1 Sol (Low)OpenAI1.93.0535
7GPT 6 Luna Non ReasoningOpenAI1.91.749
8Beam 1.0Reflection2.16.6118
9Claude Haiku 5.5 (Low)Anthropic2.26.6319
10GPT 6.1 Terra Non ReasoningOpenAI2.42.2231
11Claude Sonnet 5.5 (Low)Anthropic2.52.726
12Muse Ember (xHigh)Meta2.74.1134
13North Mini CodeCohere2.84.143
14Claude Opus 5.5 (Low)Anthropic34.7135
15Command A+Cohere3.112.636
16GPT 6 Astra (Low)OpenAI3.93.8543
17Nemotron 3 UltraNVIDIA4.39.512
18Nemotron 3.5 LightningNVIDIA4.314.196
19Mistral Medium 3.5Mistral4.624.625
20K2 Horizon 375b A23bInstitute of Foundation Models4.716.318
21Muse Glimmer (High)Meta5.523.746
22Nemotron 3.5 Super VlNVIDIA5.615.7214
23Claude Fable 5.2 (Low)Anthropic68.5554
24Inkling SmallThinking Machines6.319.8413
25DeepSeek v4.1 Flash 1017 Non ReasoningDeepSeek7.51.3819
26Muse Spark Open (xHigh)Meta7.814.2225
27GPT Oss 120b (High)OpenAI7.914.534
28Mistral Large 4 (Medium)Mistral9.823.6923
29Granite 4.2 8bIBM10.246.615
30Inkling (xHigh)Thinking Machines10.321.5112
31Muse Wonder (High)Meta10.813.141
32Grok 4.7 (Low)xAI10.910.7331
33DeepSeek v4.1 Flash 1017 (High)DeepSeek11.421.329
34Gemini 3.8 Flash (Medium)Google12.612.9130
35MiniMax M3稀宇科技13.872.8819
36Step 5 Preview阶跃星辰14.123.1129
37GLM 5.3 (Low)Zhipu14.37.1323
38Apodex 1.1Apodex14.316.2714
39Tencent Hy4Tencent14.4100.5235
40DeepSeek V4 Pro 0813 Non ReasoningDeepSeek15.32.4313
41DeepSeek V4 Pro 0813 (High)DeepSeek15.655.7721
42MiniMax m3.1 Flash稀宇科技16.453.4929
43qwen3.8 27b Non ReasoningAlibaba19.15.3610
44Gemini 4 Carbon (Low)Google2513.7254
45Agnes 2.5 Pro ThetaSapiens AI25.37.7524
46Inkling 2 SmallThinking Machines28.164.1431
47qwen3.8 27b (Low)Alibaba29.422.516
48Gemini 4 Argon (High)Google3042.8850
49qwen3.8 Flash NextAlibaba34.988.8429
50Mimo v2.6 Pro小米集团36.339.6730
51Kimi k3.1 ProMoonshot38.8169.6146
52GLM 5.3 FlashZhipu47.395.3928
53qwen3.8 Max 0902阿里巴巴50.4116.1234
54Mimo v2.6 Flash小米集团52.933.1927
55Solar Pro 4Upstage63.535.2214

数据来源

常见问题

这些分数是怎么来的?
全部来自公开测评或官方发布的机器可读数据,每一项都标注来源链接和日期。本站不编造分数,也不把缺项当成 0 分。
综合分如何计算?
先把各来源分数在该来源内部做最小-最大规范化,再按 Artificial Analysis 28%、LMArena 文本对战 28%、LiveBench 22%、编程综合 15%、推理综合 7% 加权。缺少某项时,只在已有项之间重新分配权重。至少要有两项主流测评(或三项分项)才会进入综合榜。同一系列多个变体只保留综合分最高的一条。
本站价格和评测是一回事吗?
不是。评测分数来自第三方公开榜单;人民币价格是本站模型目录的参考价(1 元 = 1,000 积分),按实际用量结算。二者分开展示,避免把计费和能力混为一谈。
多久更新一次?
采集任务每天自动运行。来源本身可能不是每天都发布新成绩,所以页面会同时给出本站采集时间和来源发布日期。