公告 AiTest123 交流群已开放 欢迎进群交流功能建议、产品想法、数据问题和使用反馈

AiTest123.comAiTest123.comAI 评测与可信度检测平台

响应速度榜

ModelSage基准跑分

分数是实测输出速度（tokens/s）的百分位，最快 = 100。官方刻意不按「相对最快模型的比例」算：榜首 813 tokens/s、第二名 401，按比例会把中间模型全压到 10 分附近。

收录模型 301 个模型 输出速度百分位 · 数据日期 2026-10-28 10:39

收录 301 个模型的输出速度百分位成绩，榜首是 Gemini 3.5 Flash-Lite（100）， 榜内平均 50.2，涉及 13 家厂商。 这份数据的时间是 2026-10-28 10:39（北京时间）。

收录模型 301 与官方榜单一致，同一模型的不同推理档位各占一行
主指标 输出速度百分位 分数是实测输出速度（tokens/s）的百分位，最快 = 100。官方刻意不按「相对最快模型的比例」算：榜首 813 tokens/s、第二名 401，按比例会把中间模型全压到 10 分附近。
榜内平均 50.2 同榜内直接可比，跨榜不可比
数据日期 2026-10-28 10:39 这一批成绩的公布时间

输出速度百分位 排名

基准跑分 · 301 个模型

名次与分数与官方榜单一致，名次不重排。条形长度按榜内最小值到最大值拉伸，领跑那条填满整列； 官方没给价格的模型显示「—」。分数后面的 ± 是官方给的统计误差，鼠标停在条上可看。

响应速度榜 输出速度百分位排名
排名模型厂商输入价（¥/M）输出价（¥/M）性价比（分/元）输出速度百分位
1 Gemini 3.5 Flash-Lite Google 2.0 16.7 17.6 100
2 Gemini 2.5 Flash-Lite Google 0.67 2.7 85.4 99.7
3 Gemini 3.1 Flash-Lite Google 1.7 10.0 26.5 99.3
4 Gemini 2.5 Flash-Lite (non-reasoning) Google 0.67 2.7 84.8 99
5 gpt-oss-120b (low) OpenAI 0.83 4.0 61.0 98.7
6 Qwen3 Coder 30B A3B 阿里百炼 1.0 4.0 56.2 98.3
7 DeepSeek V4.1 Flash (non-reasoning) DeepSeek 2.0 8.0 28.0 98
8 GPT-5.4 mini (xhigh) OpenAI 5.0 30.0 8.7 97.7
9 GPT-5.4 mini (non-reasoning) OpenAI 5.0 30.0 8.6 97.3
10 gpt-oss-120b (high) OpenAI 1.0 4.0 55.4 97
11 GPT-5.4 mini (medium) OpenAI 5.0 30.0 8.6 96.7
12 Claude Haiku 5.5 (max) Anthropic 0.67 3.3 72.2 96.3
13 Qwen3.5 Omni Flash 阿里百炼 2.2 13.3 19.3 96
14 Ministral 3 3B Mistral 0.67 0.67 143.5 95.7
15 Gemini 2.5 Flash Google 2.0 16.7 16.8 95.3
16 gpt-oss-20b (low) OpenAI 0.47 2.0 111.7 95
17 Gemini 3 Flash Google 3.3 20.0 12.6 94.7
18 Gemini 3 Flash (non-reasoning) Google 3.3 20.0 12.6 94.4
19 Qwen3.7 Max 阿里百炼 12.0 36.0 5.2 94
20 Qwen3 30B A3B 2507 (non-reasoning) 阿里百炼 0.75 3.0 71.4 93.7
21 Gemini 2.5 Flash (non-reasoning) Google 2.0 16.7 16.5 93.4
22 Gemini 3.6 Flash (high) Google 5.0 25.0 9.3 93
23 Gemini 3.5 Flash (medium) Google 10.0 60.0 4.1 92.7
24 Gemini 3.5 Flash (high) Google 10.0 60.0 4.1 92.4
25 Gemini 3.5 Flash (minimal) Google 10.0 60.0 4.1 92
26 GLM-5.2 (non-reasoning) 智谱AI 8.0 28.0 7.1 91.7
27 gpt-oss-20b (high) OpenAI 0.33 1.5 148.2 91.4
28 GPT-5 Codex (high) OpenAI 8.3 66.7 4.0 91
29 GPT-4o (Nov) OpenAI 16.7 66.7 3.1 90.7
30 GPT-4.1 OpenAI 13.3 53.3 3.9 90.4
31 Claude Haiku 5.5 (xhigh) Anthropic 0.67 3.3 67.5 90
32 Mistral Small 4 Mistral 1.0 4.0 51.2 89.7
33 DeepSeek V4.1 Flash (max) DeepSeek 2.0 8.0 25.5 89.4
34 GPT-5.4 nano (medium) OpenAI 1.3 8.3 28.9 89
35 GPT-4.1 mini OpenAI 2.7 10.7 19.0 88.7
36 GPT-5.1 Codex mini (high) OpenAI 1.7 13.3 19.3 88.4
37 Mistral Medium 3.5 Mistral 10.0 50.0 4.4 88
38 GPT-5.4 nano (xhigh) OpenAI 1.3 8.3 28.4 87.7
39 GPT-5.4 nano (non-reasoning) OpenAI 1.3 8.3 28.3 87.4
40 Qwen3.5 35B A3B (non-reasoning) 阿里百炼 0.40 3.2 79.1 87
41 Mistral Small 4 (non-reasoning) Mistral 1.0 4.0 49.5 86.7
42 DeepSeek V4 Flash Vision (max) DeepSeek 2.9 8.8 19.6 86.4
43 DeepSeek R1 (Jan) DeepSeek 4.0 16.0 12.3 86
44 GPT-5 nano (high) OpenAI 0.33 2.7 93.5 85.7
45 GPT-5 nano (medium) OpenAI 0.33 2.7 93.1 85.4
46 DeepSeek V4.1 Flash 1017 (max) DeepSeek 2.0 8.0 24.3 85
47 GPT-5.5 Instant (June 2026) OpenAI 33.3 200 1.1 84.7
48 Claude Haiku 5.5 (high) Anthropic 0.67 3.3 63.3 84.4
49 GPT-5 nano (minimal) OpenAI 0.33 2.7 91.7 84.1
50 o3 OpenAI 13.3 53.3 3.6 83.7
51 Qwen3.5 35B A3B 阿里百炼 0.40 3.2 75.8 83.4
52 Claude Haiku 5.5 (low) Anthropic 0.67 3.3 62.3 83.1
53 Claude Haiku 5.5 (medium) Anthropic 0.67 3.3 62.0 82.7
54 DeepSeek V4 Flash 0731 (max) DeepSeek 2.0 8.0 23.5 82.4
55 Qwen3.6 35B A3B (non-reasoning) 阿里百炼 1.8 10.8 20.3 82.1
56 Qwen3.5 122B A10B (non-reasoning) 阿里百炼 0.80 6.4 37.1 81.7
57 Qwen3 Next 80B A3B (指令) 阿里百炼 1.0 4.0 46.5 81.4
58 Gemma 4 12B (non-reasoning) Google 0.67 2.0 81.1 81.1
59 Claude Sonnet 5.5 (max with fallback) Anthropic 13.3 66.7 3.0 80.7
60 GPT-4o (Aug) OpenAI 16.7 66.7 2.8 80.4
61 GPT-5.1 (high) OpenAI 8.3 66.7 3.5 80.1
62 Ministral 3 8B Mistral 1.0 1.0 79.7 79.7
63 Gemma 4 12B Google 0.67 2.0 79.4 79.4
64 Qwen3.6 35B A3B 阿里百炼 1.8 10.8 19.5 79.1
65 DeepSeek V3 0324 DeepSeek 2.0 8.0 22.5 78.7
66 Qwen3.5 122B A10B 阿里百炼 0.80 6.4 35.6 78.4
67 GPT-5.2 Codex (xhigh) OpenAI 11.7 93.4 2.4 78.1
68 GPT-5.6 Terra (max) OpenAI 13.3 80.0 2.6 77.7
69 Gemini 3.1 Pro Preview Google 13.3 80.0 2.6 77.4
70 Qwen3 30B 阿里百炼 0.75 7.5 31.6 77.1
71 GPT-5.1 Codex (high) OpenAI 8.3 66.7 3.3 76.7
72 GPT-6 Sol (max) OpenAI 13.3 66.7 2.9 76.4
73 GPT-5.6 Luna (max) OpenAI 1.3 8.0 25.4 76.1
74 Gemini 3.8 Flash (high) Google 5.0 25.0 7.6 75.7
75 Grok 4.20 0309 v2 xAI 8.3 16.7 7.2 75.4
76 Kimi K2 Thinking Moonshot/Kimi 4.0 16.0 10.7 75.1
77 GPT-6 Sol (low) OpenAI 13.3 66.7 2.8 74.8
78 GPT-5.6 Luna (xhigh) OpenAI 1.3 8.0 24.8 74.4
79 GPT-6 Luna (max) OpenAI 0.67 3.3 55.6 74.1
80 DeepSeek V3.1 DeepSeek 4.0 12.0 12.3 73.8
81 GPT-5.6 Luna (high) OpenAI 1.3 8.0 24.5 73.4
82 GPT-5.4 (xhigh) OpenAI 16.7 100 1.9 73.1
83 Qwen3 30B (non-reasoning) 阿里百炼 0.75 3.0 55.5 72.8
84 GPT-4o (May) OpenAI 33.3 100 1.4 72.4
85 GPT-5.1 (non-reasoning) OpenAI 8.3 66.7 3.1 72.1
86 GPT-6 Sol (high) OpenAI 13.3 66.7 2.7 71.8
87 GPT-6 Luna (high) OpenAI 0.67 3.3 53.5 71.4
88 GPT-6 Luna (xhigh) OpenAI 0.67 3.3 53.3 71.1
89 Mistral Large 4 Preview Mistral 9.1 27.9 5.1 70.8
90 GPT-5.6 Luna (low) OpenAI 1.3 8.0 23.5 70.4
91 GPT-6 Sol (xhigh) OpenAI 13.3 66.7 2.6 70.1
92 GPT-5.6 Luna (non-reasoning) OpenAI 1.3 8.0 23.3 69.8
93 Qwen3.5 Omni Plus 阿里百炼 7.0 40.0 4.6 69.4
94 Qwen3 235B A22B 2507 阿里百炼 2.0 20.0 10.6 69.1
95 Gemma 4 26B A4B (non-reasoning) Google 0.77 2.7 55.4 68.8
96 GPT-5.6 Luna (medium) OpenAI 1.3 8.0 22.8 68.4
97 GPT-5 (high) OpenAI 8.3 66.7 3.0 68.1
98 GLM-4.7-Flash (non-reasoning)FREE 智谱AI 免费 免费 6780.0 67.8
99 GPT-6 Sol (non-reasoning) OpenAI 13.3 66.7 2.5 67.4
100 Gemini 2.5 Pro Google 8.3 66.7 2.9 67.1
101 Qwen3 Omni 30B A3B (推理) 阿里百炼 1.8 12.7 14.8 66.8
102 GPT-6 Luna (low) OpenAI 0.67 3.3 49.8 66.4
103 GPT-5 mini (medium) OpenAI 1.7 13.3 14.4 66.1
104 GPT-5.6 Terra (high) OpenAI 13.3 80.0 2.2 65.8
105 MiniMax M1 80k MiniMax 2.1 8.4 17.8 65.4
106 GPT-5.6 Terra (non-reasoning) OpenAI 13.3 80.0 2.2 65.1
107 Qwen3 Omni 30B A3B (指令) 阿里百炼 1.8 12.7 14.3 64.8
108 GPT-6 Luna (non-reasoning) OpenAI 0.67 3.3 48.4 64.5
109 GPT-5.6 Terra (low) OpenAI 13.3 80.0 2.1 64.1
110 DeepSeek V4 Pro (non-reasoning) DeepSeek 9.0 27.0 4.7 63.8
111 MiniMax-M3.1-Flash MiniMax 1.0 4.0 36.3 63.5
112 GPT-5.6 Terra (medium) OpenAI 13.3 80.0 2.1 63.1
113 Grok 4.20 0309 v2 (non-reasoning) xAI 8.3 16.7 6.0 62.8
114 GPT-5 mini (minimal) OpenAI 1.7 13.3 13.6 62.5
115 GPT-5 mini (high) OpenAI 1.7 13.3 13.5 62.1
116 DeepSeek V4 Pro (high) DeepSeek 9.0 27.0 4.6 61.8
117 DeepSeek V4 Pro (max) DeepSeek 9.0 27.0 4.6 61.5
118 Grok 4.3 (medium) xAI 8.3 16.7 5.9 61.1
119 GPT-4o mini OpenAI 1.0 4.0 34.7 60.8
120 GPT-5.6 Terra (xhigh) OpenAI 13.3 80.0 2.0 60.5
121 GPT-5.3 Codex (xhigh) OpenAI 11.7 93.4 1.9 60.1
122 MiniMax-M2 MiniMax 2.1 8.4 16.3 59.8
123 GPT-5 (medium) OpenAI 8.3 66.7 2.6 59.5
124 GPT-5 (low) OpenAI 8.3 66.7 2.6 59.1
125 Gemma 4 26B A4B Google 0.67 2.3 55.1 58.8
126 Kimi K2.6 Moonshot/Kimi 6.5 27.0 5.0 58.5
127 DeepSeek V3.1 (non-reasoning) DeepSeek 4.0 12.0 9.7 58.1
128 Grok 4.3 (non-reasoning) xAI 8.3 16.7 5.5 57.8
129 Qwen3 Coder Next 阿里百炼 1.0 4.0 32.9 57.5
130 MiniMax-M3 MiniMax 4.2 16.8 7.8 57.1
131 Claude Opus 5.5 (max with fallback) Anthropic 26.7 133 1.1 56.8
132 Kimi K2.6 (non-reasoning) Moonshot/Kimi 6.5 27.0 4.9 56.5
133 Grok 4.3 xAI 8.3 16.7 5.4 56.1
134 DeepSeek V4 Pro 0813 (max) DeepSeek 9.0 27.0 4.1 55.8
135 Grok Build 0.1 0616 xAI 6.7 13.3 6.7 55.5
136 Claude Sonnet 5.5 (xhigh with fallback) Anthropic 13.3 66.7 2.1 55.1
137 GLM-4.6V (non-reasoning) 智谱AI 1.0 3.0 36.5 54.8
138 DeepSeek V4 Pro 0813 (non-reasoning) DeepSeek 9.0 27.0 4.0 54.5
139 Claude 4.5 Haiku Anthropic 6.7 33.3 4.1 54.2
140 GLM-5.2 (max) 智谱AI 8.0 28.0 4.1 53.8
141 GPT-5.5 (xhigh) OpenAI 33.3 200 0.7 53.5
142 GPT-6.1 Terra (non-reasoning) OpenAI 6.7 33.3 4.0 53.2
143 GPT-6.1 Terra (high) OpenAI 6.7 33.3 4.0 52.8
144 GPT-6.1 Terra (xhigh) OpenAI 6.7 33.3 3.9 52.5
145 GPT-6.1 Terra (max) OpenAI 6.7 33.3 3.9 52.2
146 GPT-6.1 Terra (medium) OpenAI 6.7 33.3 3.9 51.8
147 GPT-6.1 Terra (low) OpenAI 6.7 33.3 3.9 51.5
148 Claude Sonnet 5.5 (high with fallback) Anthropic 13.3 66.7 1.9 51.2
149 GPT-5 (minimal) OpenAI 8.3 66.7 2.2 50.8
150 Mistral Large 3 Mistral 3.3 10.0 10.1 50.5
151 Claude Sonnet 5.5 (low with fallback) Anthropic 13.3 66.7 1.9 50.2
152 GLM-4.7-FlashFREE 智谱AI 免费 免费 4980.0 49.8
153 Claude Sonnet 5.5 (medium with fallback) Anthropic 13.3 66.7 1.9 49.5
154 Claude 4.5 Haiku (non-reasoning) Anthropic 6.7 33.3 3.7 49.2
155 Grok 3 xAI 20.0 100 1.2 48.8
156 Qwen3 Coder 480B 阿里百炼 6.0 24.0 4.6 48.5
157 GLM-4.5-Air 智谱AI 0.80 2.0 43.8 48.2
158 GLM-4.6V 智谱AI 1.0 3.0 31.9 47.8
159 GLM-5.3 (max) 智谱AI 8.0 28.0 3.7 47.5
160 GPT-5.6 Sol (xhigh) OpenAI 26.7 133 0.9 47.2
161 GPT-5.5 (high) OpenAI 33.3 200 0.6 46.8
162 GPT-5.6 Sol (high) OpenAI 26.7 133 0.9 46.5
163 Claude Sonnet 5 (max) Anthropic 13.3 66.7 1.7 46.2
164 GPT-5.5 (medium) OpenAI 33.3 200 0.6 45.8
165 GPT-5.5 (low) OpenAI 33.3 200 0.6 45.5
166 GPT-5.6 Sol (max) OpenAI 26.7 133 0.8 45.2
167 Qwen3.5 27B (non-reasoning) 阿里百炼 0.60 4.8 27.2 44.9
168 GPT-5.2 (xhigh) OpenAI 11.7 93.4 1.4 44.5
169 Kimi K2.5 Moonshot/Kimi 4.0 21.0 5.4 44.2
170 Qwen3.5 9B (non-reasoning) 阿里百炼 0.50 2.0 50.2 43.9
171 Hy3 腾讯混元 0.91 3.7 27.1 43.5
172 GPT-5.5 (non-reasoning) OpenAI 33.3 200 0.6 43.2
173 Grok 4.6 (xhigh) xAI 13.3 40.0 2.1 42.9
174 Grok 4.3 (low) xAI 8.3 16.7 4.1 42.5
175 GPT-5.6 Sol (low) OpenAI 26.7 133 0.8 42.2
176 Ministral 3 14B Mistral 1.3 1.3 31.4 41.9
177 Magistral Small 1.2 Mistral 3.3 10.0 8.3 41.5
178 Grok 4 xAI 20.0 100 1.0 41.2
179 Gemini 4 Argon (high) Google 13.3 66.7 1.5 40.9
180 Claude Opus 5.5 (xhigh with fallback) Anthropic 26.7 133 0.8 40.5
181 Qwen3.5 27B 阿里百炼 0.60 4.8 24.4 40.2
182 Grok 4.7 (high) xAI 13.3 40.0 2.0 39.9
183 Grok 4.6 (high) xAI 13.3 40.0 2.0 39.5
184 Grok 4.7 (xhigh) xAI 13.3 40.0 2.0 39.2
185 DeepSeek V3.1 Terminus (non-reasoning) DeepSeek 4.0 12.0 6.5 38.9
186 GPT-5.6 Sol (medium) OpenAI 26.7 133 0.7 38.5
187 Kimi K2.7 Code Moonshot/Kimi 6.5 27.0 3.3 38.2
188 GPT-5.4 (non-reasoning) OpenAI 16.7 100 1.0 37.9
189 GLM-5.1 智谱AI 6.0 24.0 3.6 37.5
190 GLM-5.3 (low) 智谱AI 8.0 28.0 2.9 37.2
191 Grok 4.6 (low) xAI 13.3 40.0 1.8 36.9
192 Qwen3 14B 阿里百炼 1.0 10.0 11.2 36.5
193 GPT-5.4 (low) OpenAI 16.7 100 1.0 36.2
194 GPT-5.6 Sol (non-reasoning) OpenAI 26.7 133 0.7 35.9
195 GLM-5.1 (non-reasoning) 智谱AI 6.0 24.0 3.4 35.5
196 Claude Opus 5.5 (high with fallback) Anthropic 26.7 133 0.7 35.2
197 Grok 4.6 (medium) xAI 13.3 40.0 1.7 34.9
198 GLM-5 (non-reasoning) 智谱AI 4.0 18.0 4.6 34.6
199 GLM-5 智谱AI 4.0 18.0 4.6 34.2
200 Grok 4.7 (low) xAI 13.3 40.0 1.7 33.9
201 Claude Opus 5.5 (low with fallback) Anthropic 26.7 133 0.6 33.6
202 Qwen3 14B (non-reasoning) 阿里百炼 1.0 4.0 19.0 33.2
203 Claude Opus 5.5 (medium with fallback) Anthropic 26.7 133 0.6 32.9
204 GLM-4.5V (non-reasoning) 智谱AI 2.0 6.0 10.9 32.6
205 Claude Sonnet 5 (medium) Anthropic 13.3 66.7 1.2 32.2
206 GPT-5.2 (non-reasoning) OpenAI 11.7 93.4 1.0 31.9
207 Claude Sonnet 5 (xhigh) Anthropic 13.3 66.7 1.2 31.6
208 GLM-5.3-Flash 智谱AI 0.80 2.8 24.0 31.2
209 Claude Fable 5 (with fallback) Anthropic 66.7 333 0.2 30.9
210 Claude Sonnet 5 (non-reasoning) Anthropic 13.3 66.7 1.1 30.6
211 Claude Sonnet 5 (low) Anthropic 13.3 66.7 1.1 30.2
212 Claude Sonnet 5 (high) Anthropic 13.3 66.7 1.1 29.9
213 Claude Fable 5.1 (max with fallback) Anthropic 66.7 333 0.2 29.6
214 GLM-4.5V 智谱AI 2.0 6.0 9.7 29.2
215 Grok 4.5 (high) xAI 13.3 40.0 1.4 28.9
216 Gemma 4 31B (non-reasoning) Google 1.1 2.7 19.2 28.6
217 DeepSeek V4 Flash (max) DeepSeek 2.0 8.0 8.1 28.2
218 Qwen3 32B (non-reasoning) 阿里百炼 2.0 8.0 8.0 27.9
219 GPT-6 Astra (max) OpenAI 66.7 333 0.2 27.6
220 Claude Sonnet 4.6 (max) Anthropic 20.0 100 0.7 27.2
221 Claude Opus 4.8 (max) Anthropic 33.3 167 0.4 26.9
222 Qwen3.5 9B 阿里百炼 0.50 2.0 30.4 26.6
223 GPT-6.1 Sol (max) OpenAI 13.3 66.7 1.0 26.2
224 Qwen3.5 397B A17B 阿里百炼 1.2 7.2 9.6 25.9
225 Qwen3.5 397B A17B (non-reasoning) 阿里百炼 1.2 7.2 9.5 25.6
226 MiniMax-M2.1 MiniMax 2.1 8.4 6.9 25.2
227 Qwen3.6 27B (non-reasoning) 阿里百炼 3.0 18.0 3.7 24.9
228 MiniMax-M2.5 MiniMax 2.1 8.4 6.7 24.6
229 GPT-6 Astra (low) OpenAI 66.7 333 0.2 24.3
230 Qwen3.6 27B 阿里百炼 3.0 18.0 3.5 23.9
231 MiniMax-M2.7 MiniMax 2.1 8.4 6.4 23.6
232 Qwen3.7 Plus 阿里百炼 2.0 8.0 6.7 23.3
233 GPT-6.1 Sol (xhigh) OpenAI 13.3 66.7 0.9 22.9
234 GPT-6.1 Sol (low) OpenAI 13.3 66.7 0.8 22.6
235 Qwen3.6 Plus 阿里百炼 2.0 12.0 5.0 22.3
236 Claude Opus 4.5 Anthropic 33.3 167 0.3 21.9
237 Kimi K3 (max) Moonshot/Kimi 20.0 100 0.5 21.6
238 GPT-6 Astra (xhigh) OpenAI 66.7 333 0.2 21.3
239 Gemma 4 E4B (non-reasoning) Google 0.13 0.67 78.4 20.9
240 Gemma 4 E4B Google 0.13 0.67 77.2 20.6
241 Mistral Small 3 Mistral 0.33 0.53 52.9 20.3
242 GPT-6.1 Sol (high) OpenAI 13.3 66.7 0.7 19.9
243 Claude Opus 5 (high) Anthropic 33.3 167 0.3 19.6
244 Kimi K3 (low) Moonshot/Kimi 20.0 100 0.5 19.3
245 Claude Fable 5.1 (xhigh with fallback) Anthropic 66.7 333 0.1 18.9
246 Qwen3.8-Flash-Next 阿里百炼 1.0 3.1 12.1 18.6
247 MiMo-V2.6-Flash 小米 0.93 1.9 15.7 18.3
248 Claude Opus 5 (low) Anthropic 33.3 167 0.3 17.9
249 Claude Opus 5 (xhigh) Anthropic 33.3 167 0.3 17.6
250 GPT-6 Astra (high) OpenAI 66.7 333 0.1 17.3
251 Claude Opus 5 (max) Anthropic 33.3 167 0.3 16.9
252 Kimi K2.5 (non-reasoning) Moonshot/Kimi 4.0 21.0 2.0 16.6
253 Mistral Small 3.2 Mistral 0.50 1.3 23.0 16.3
254 Claude Opus 4.7 (max) Anthropic 33.3 167 0.2 15.9
255 MiMo-V2.5-Pro (non-reasoning) 小米 2.6 7.8 4.0 15.6
256 GPT-6 Astra (medium) OpenAI 66.7 333 0.1 15.3
257 MiMo-V2.5-Pro 小米 2.7 6.8 4.0 15
258 Grok 3 mini Reasoning (high) xAI 2.0 3.3 6.3 14.6
259 GPT-6.1 Sol (medium) OpenAI 13.3 66.7 0.5 14.3
260 Mistral Small 3.1 Mistral 0.53 0.90 22.4 14
261 Claude 4.5 Sonnet Anthropic 20.0 100 0.3 13.6
262 Claude Sonnet 4.6 (non-reasoning, low) Anthropic 20.0 100 0.3 13.3
263 Claude Opus 4.5 (non-reasoning) Anthropic 33.3 167 0.2 13
264 Claude Opus 5 (medium) Anthropic 33.3 167 0.2 12.6
265 Claude Sonnet 4.6 (non-reasoning, high) Anthropic 20.0 100 0.3 12.3
266 Qwen3 235B 2507 阿里百炼 2.0 8.0 3.4 12
267 Claude Fable 5.1 (low with fallback) Anthropic 66.7 333 0.1 11.6
268 Kimi K2 0905 Moonshot/Kimi 4.0 16.0 1.6 11.3
269 Claude Opus 4.7 (non-reasoning, high) Anthropic 33.3 167 0.2 11
270 Qwen3.8 27B (low) 阿里百炼 3.0 12.0 2.0 10.6
271 Kimi K2 Moonshot/Kimi 4.0 16.0 1.5 10.3
272 Claude Fable 5.1 (high with fallback) Anthropic 66.7 333 0.1 10
273 Claude 4.5 Sonnet (non-reasoning) Anthropic 20.0 100 0.2 9.6
274 Claude Fable 5.1 (medium with fallback) Anthropic 66.7 333 0.1 9.3
275 Mistral Medium 3 Mistral 2.7 13.3 1.7 9
276 Qwen3 VL 235B A22B (指令) 阿里百炼 1.0 10.0 2.6 8.6
277 Qwen3.8 27B (non-reasoning) 阿里百炼 3.0 12.0 1.6 8.3
278 Kimi K3.1 Pro Moonshot/Kimi 13.0 68.0 0.3 8
279 MiMo-V2.6-Pro 小米 2.9 5.8 2.1 7.6
280 Qwen3.8 27B (medium) 阿里百炼 3.0 12.0 1.4 7.3
281 DeepSeek V3 (Dec) DeepSeek 2.0 8.0 2.0 7
282 Claude Opus 4.6 (max) Anthropic 33.3 167 0.1 6.6
283 Claude Opus 4.6 (non-reasoning, high) Anthropic 33.3 167 0.1 6.3
284 Qwen3.8 Max 阿里百炼 12.0 36.0 0.3 6
285 Qwen3.8 27B (xhigh) 阿里百炼 3.0 12.0 1.1 5.6
286 Qwen3.8 2.4T A95B 阿里百炼 12.0 36.0 0.3 5.3
287 Qwen3.8 Max (0902) 阿里百炼 12.0 36.0 0.3 5
288 Seed-OSS-36B-Instruct 火山引擎/豆包 0.60 3.6 3.5 4.7
289 MiMo-V2.5 小米 0.89 1.8 3.9 4.3
290 GLM-4.7 智谱AI 2.0 8.0 1.1 4
291 GLM-4.7 (non-reasoning) 智谱AI 2.0 8.0 1.1 3.7
292 GLM-4.6 (non-reasoning) 智谱AI 1.0 3.0 2.2 3.3
293 DeepSeek R1 0528 DeepSeek 4.0 16.0 0.4 3
294 Gemma 4 31BFREE Google 免费 免费 270.0 2.7
295 GLM-4.6 智谱AI 1.0 3.0 1.5 2.3
296 Qwen3 32B 阿里百炼 2.0 20.0 0.3 2
297 Qwen3 VL 30B A3B (指令) 阿里百炼 0.15 1.5 3.5 1.7
298 Qwen2.5 72B 阿里百炼 2.0 8.0 0.4 1.3
299 Mixtral 8x7B Mistral 3.0 4.7 0.3 1
300 Qwen3.5 4B 阿里百炼 0.30 1.2 1.3 0.7
301 Qwen3.5 4B (non-reasoning) 阿里百炼 0.30 1.2 0.6 0.3

评分方式与读法

这张榜怎么看
做实时对话、补全这类等不起的场景先过一遍这张。
收录模型301与官方榜单一致，同一模型的不同推理档位各占一行
主指标输出速度百分位分数是实测输出速度（tokens/s）的百分位，最快 = 100。官方刻意不按「相对最快模型的比例」算：榜首 813 tokens/s、第二名 401，按比例会把中间模型全压到 10 分附近。
成绩时间2026-10-28 10:39这份成绩的日期（北京时间），官方停更时这里也停住
评分方式按官方原样列出分数、价格、名次都没有二次计算；官方调整评分方式时这里跟着变
同榜怎么看先看价格再看分数先按能接受的价格筛，再在筛出的里面挑分高的，比只看榜首实用
跨榜怎么用别比大小，只看排序基准榜 0-100 与盲测榜 Elo 量纲不同；跨榜只参考「排在第几档」

成绩取自 ModelSage 榜单页（https://modelsage.cn/leaderboards/speed），本站只取回与整理，不改算。 做实时对话、补全这类等不起的场景先过一遍这张。

其它榜

16 张
全部榜 →

综合能力榜

AA 智能指数相对分

393 个模型

代码能力榜

Terminal-Bench + SciCode 相对分

177 个模型

Agent 任务榜

Terminal-Bench 相对分

177 个模型

性价比榜

分/元

78 个模型

Agent 竞技总榜

5 维行为信号加权分

55 个模型

文本对话榜

Elo 分

418 个模型

搜索问答榜

Elo 分

42 个模型

视觉理解榜

Elo 分

167 个模型

文档理解榜

Elo 分

44 个模型

代码竞技榜

Elo 分

151 个模型

图生代码榜

Elo 分

65 个模型

文生图榜

Elo 分

82 个模型

图像编辑榜

Elo 分

60 个模型

文生视频榜

Elo 分

48 个模型

图生视频榜

Elo 分

48 个模型

视频编辑榜

Elo 分

10 个模型

成绩来自 ModelSage 公开榜单，数据日期 2026-10-28 10:39。