M
ModelSage
排行榜/响应速度

AI响应速度排行榜

基于AA实测输出速度(tokens/s)百分位排名，找到最快的大模型。

分数怎么算：分数是实测输出速度的百分位（最快 = 100 分）。不按「相对最快模型的比例」算：榜首 813 tokens/s、第二名 401，按比例算会把中间的模型全压到 10 分附近。

数据更新于 2026-10-14，每天北京时间凌晨自动刷新

综合能力代码能力Agent 智能体Arena 人类偏好视觉/多模态搜索能力文档理解响应速度性价比
RankModelScore
1
Gemini 3.5 Flash-Lite
100.0
2
Gemini 3.1 Flash-Lite
99.7
3
Gemini 3.7 Flash (low)
99.3
4
Gemini 2.5 Flash-Lite
99.0
5
Gemini 3.7 Flash (high)
98.7
6
Gemini 3.7 Flash (medium)
98.3
7
gpt-oss-20b (low)
98.0
8
Gemini 2.5 Flash-Lite (non-reasoning)
97.7
9
gpt-oss-120b (low)
97.3
10
Claude Haiku 5.5 (max)
97.0
11
GPT-5.4 mini (xhigh)
96.7
12
gpt-oss-120b (high)
96.3
13
Qwen3.5 Omni Flash
96.0
14
GPT-4.1 nano
95.7
15
gpt-oss-20b (high)
95.3
16
Ministral 3 3B
95.0
17
Gemini 2.5 Flash
94.7
18
Gemini 2.5 Flash (non-reasoning)
94.4
19
Gemini 3 Flash (non-reasoning)
94.0
20
GPT-5.4 mini (medium)
93.7
21
GPT-5 nano (high)
93.4
22
o3-mini (high)
93.0
23
Qwen3.7 Max
92.7
24
GPT-5.4 mini (non-reasoning)
92.4
25
Gemini 3.5 Flash (medium)
92.0
26
Gemini 3.6 Flash (high)
91.7
27
DeepSeek V4.1 Flash (non-reasoning)
91.4
28
Gemini 3 Flash
91.0
29
GPT-5 nano (minimal)
90.7
30
Gemini 3.5 Flash (high)
90.4
31
o3-mini
90.0
32
DeepSeek V4 Flash Vision (max)
89.7
33
DeepSeek V4.1 Flash (max)
89.4
34
DeepSeek V4 Flash 0731 (max)
89.0
35
Claude Haiku 5.5 (xhigh)
88.7
36
GPT-5 nano (medium)
88.4
37
Gemini 3.5 Flash (minimal)
88.0
38
GPT-5.4 nano (medium)
87.7
39
GPT-5 Codex (high)
87.4
40
Claude Haiku 5.5 (medium)
87.0
41
Claude Haiku 5.5 (high)
86.7
42
GPT-5.4 nano (xhigh)
86.4
43
Claude Haiku 5.5 (low)
86.0
44
GPT-5.4 nano (non-reasoning)
85.7
45
Mistral Medium 3.5
85.4
46
Mistral Small 4
85.0
47
Qwen3.5 35B A3B (non-reasoning)
84.7
48
GPT-5.1 Codex mini (high)
84.4
49
Qwen3 Next 80B A3B (指令)
84.1
50
Kimi K2 Thinking
83.7
51
Qwen3 Coder 30B A3B
83.4
52
GPT-5.5 Instant (June 2026)
83.1
53
Qwen3 Next 80B A3B (推理)
82.7
54
GPT-5 mini (high)
82.4
55
GPT-4.1 mini
82.1
56
GPT-5 mini (minimal)
81.7
57
GPT-5 mini (medium)
81.4
58
o4-mini (high)
81.1
59
GLM-5.2 (non-reasoning)
80.7
60
DeepSeek R1 (Jan)
80.4
61
GPT-4.1
80.1
62
Qwen3.5 35B A3B
79.7
63
GPT-4o (Nov)
79.4
64
Mistral Small 4 (non-reasoning)
79.1
65
Gemini 3.8 Flash (high)
78.7
66
GPT-4o (Aug)
78.4
67
Qwen3.6 35B A3B (non-reasoning)
78.1
68
Qwen3.5 122B A10B (non-reasoning)
77.7
69
Gemini 3.1 Pro Preview
77.4
70
GPT-4o (May)
77.1
71
Claude Sonnet 5.5 (max with fallback)
76.7
72
GPT-5.2 Codex (xhigh)
76.4
73
GPT-6 Luna (max)
76.1
74
GPT-5.6 Terra (max)
75.7
75
Qwen3.6 35B A3B
75.4
76
GPT-6 Luna (high)
75.1
77
Qwen3.5 122B A10B
74.8
78
GPT-6 Luna (xhigh)
74.4
79
GPT-5.1 Codex (high)
74.1
80
GPT-5.6 Luna (max)
73.8
81
GPT-6 Luna (low)
73.4
82
o3
73.1
83
DeepSeek V3.1
72.8
84
GPT-4o mini
72.4
85
GPT-6 Sol (max)
72.1
86
GLM-4.7-Flash (non-reasoning)FREE
71.8
87
GPT-5.6 Luna (medium)
71.4
88
GPT-5.6 Luna (xhigh)
71.1
89
GPT-5.1 (high)
70.8
90
GPT-5.6 Luna (high)
70.4
91
Gemini 2.5 Pro
70.1
92
Gemma 4 12B (non-reasoning)
69.8
93
Grok 4.3
69.4
94
GPT-6 Sol (xhigh)
69.1
95
Qwen3 Coder 480B
68.8
96
GPT-6 Luna (non-reasoning)
68.4
97
GPT-5.6 Luna (non-reasoning)
68.1
98
GPT-5.3 Codex (xhigh)
67.8
99
Gemma 4 12B
67.4
100
GPT-6 Sol (high)
67.1
101
GPT-5 (low)
66.8
102
GPT-5.6 Terra (xhigh)
66.4
103
GPT-5.6 Luna (low)
66.1
104
GPT-5.6 Terra (high)
65.8
105
Grok 3
65.4
106
MiniMax-M3
65.1
107
GPT-6 Sol (non-reasoning)
64.8
108
GPT-5.1 (non-reasoning)
64.5
109
GPT-5.6 Terra (medium)
64.1
110
Ministral 3 8B
63.8
111
GPT-5.6 Terra (non-reasoning)
63.5
112
GPT-5.6 Terra (low)
63.1
113
Claude Sonnet 5.5 (xhigh with fallback)
62.8
114
Qwen3 Omni 30B A3B (推理)
62.5
115
Qwen3 30B
62.1
116
GPT-5 (high)
61.8
117
GPT-5.5 (xhigh)
61.5
118
DeepSeek V4 Pro (non-reasoning)
61.1
119
Claude Sonnet 5.5 (low with fallback)
60.8
120
DeepSeek V4 Pro (max)
60.5
121
Grok 4.3 (medium)
60.1
122
DeepSeek V4 Pro 0813 (max)
59.8
123
Qwen3.5 9B (non-reasoning)
59.5
124
GPT-6 Sol (low)
59.1
125
DeepSeek V4 Pro 0813 (non-reasoning)
58.8
126
Claude Sonnet 5.5 (high with fallback)
58.5
127
Qwen3 Omni 30B A3B (指令)
58.1
128
Gemma 4 26B A4B (non-reasoning)
57.8
129
Grok 4.20 0309 v2
57.5
130
GPT-5 (medium)
57.1
131
DeepSeek V4 Pro (high)
56.8
132
Claude Sonnet 5.5 (medium with fallback)
56.5
133
Claude Opus 5.5 (max with fallback)
56.1
134
Grok 4.3 (low)
55.8
135
DeepSeek V3.1 (non-reasoning)
55.5
136
Grok 4.3 (non-reasoning)
55.1
137
Claude 4.5 Haiku
54.8
138
GPT-5.6 Sol (max)
54.5
139
GPT-5.5 (high)
54.2
140
GPT-5.4 (xhigh)
53.8
141
Grok 4.20 0309 v2 (non-reasoning)
53.5
142
GPT-5.5 (medium)
53.2
143
Qwen3.5 Omni Plus
52.8
144
Qwen3 30B (non-reasoning)
52.5
145
Claude 4.5 Haiku (non-reasoning)
52.2
146
GLM-4.6V
51.8
147
GPT-5 (minimal)
51.5
148
GPT-5.5 (non-reasoning)
51.2
149
GPT-5.6 Sol (xhigh)
50.8
150
Qwen3 235B A22B 2507
50.5
151
GPT-5.6 Sol (high)
50.2
152
Grok 4.7 (xhigh)
49.8
153
GPT-5.5 (low)
49.5
154
Qwen2.5 Turbo
49.2
155
Hy3
48.8
156
Claude Sonnet 5 (max)
48.5
157
Claude Opus 5.5 (high with fallback)
48.2
158
GLM-5.2 (max)
47.8
159
Claude Opus 5.5 (xhigh with fallback)
47.5
160
DeepSeek V3 0324
47.2
161
Qwen3.5 27B (non-reasoning)
46.8
162
GPT-5.2 (xhigh)
46.5
163
GPT-5.6 Sol (medium)
46.2
164
Claude Opus 5.5 (low with fallback)
45.8
165
Claude Opus 5.5 (medium with fallback)
45.5
166
Qwen3.5 27B
45.2
167
Grok 4.7 (high)
44.9
168
GPT-5.6 Sol (non-reasoning)
44.5
169
GPT-5.6 Sol (low)
44.2
170
GLM-4.5-Air
43.9
171
GLM-5
43.5
172
Magistral Small 1.2
43.2
173
Grok 4
42.9
174
GLM-5.3 (low)
42.5
175
Mistral Large 3
42.2
176
Qwen3 32B (non-reasoning)
41.9
177
MiniMax-M2
41.5
178
GPT-5.2 (non-reasoning)
41.2
179
GLM-4.6V (non-reasoning)
40.9
180
GLM-5 (non-reasoning)
40.5
181
Claude Fable 5.1 (max with fallback)
40.2
182
Qwen3 Coder Next
39.9
183
Claude Sonnet 5 (xhigh)
39.5
184
Gemma 4 26B A4B
39.2
185
GLM-5.1 (non-reasoning)
38.9
186
MiniMax-M2.5
38.5
187
GLM-4.7-FlashFREE
38.2
188
MiniMax M1 80k
37.9
189
Grok Build 0.1 0616
37.5
190
Grok 3 mini Reasoning (high)
37.2
191
GPT-5.4 (non-reasoning)
36.9
192
GLM-5.3-Flash
36.5
193
MiniMax-M2.1
36.2
194
Claude Sonnet 5 (high)
35.9
195
Claude Fable 5 (with fallback)
35.5
196
Qwen3.5 9B
35.2
197
GLM-5.1
34.9
198
Claude Sonnet 5 (medium)
34.6
199
Mistral Small 3
34.2
200
MiMo-V2.6-Flash
33.9
201
Grok 4.7 (low)
33.6
202
Qwen3.5 397B A17B
33.2
203
Claude Sonnet 5 (non-reasoning)
32.9
204
MiniMax-M2.7
32.6
205
Kimi K2.7 Code
32.2
206
Claude Fable 5.1 (xhigh with fallback)
31.9
207
Grok 4.5 (high)
31.6
208
Qwen3.5 397B A17B (non-reasoning)
31.2
209
GLM-5.3 (max)
30.9
210
GPT-5.4 (low)
30.6
211
GPT-6 Astra (max)
30.2
212
Claude Opus 4.8 (max)
29.9
213
Grok 4.6 (xhigh)
29.6
214
Claude Sonnet 5 (low)
29.2
215
Claude Sonnet 4.6 (max)
28.9
216
Kimi K2.5
28.6
217
Kimi K2.5 (non-reasoning)
28.2
218
GPT-6.1 Sol (xhigh)
27.9
219
GPT-6.1 Sol (max)
27.6
220
Claude Fable 5.1 (low with fallback)
27.2
221
Qwen3.6 27B
26.9
222
Kimi K2 0905
26.6
223
GPT-6 Astra (xhigh)
26.2
224
Grok 4.6 (high)
25.9
225
Kimi K2
25.6
226
GPT-6.1 Sol (high)
25.2
227
Qwen3.6 27B (non-reasoning)
24.9
228
Claude Opus 5 (max)
24.6
229
Grok 4.6 (medium)
24.3
230
Qwen3.6 Plus
23.9
231
Qwen3.7 Plus
23.6
232
Claude Opus 5 (low)
23.3
233
Claude Fable 5.1 (medium with fallback)
22.9
234
GPT-6.1 Sol (medium)
22.6
235
Ministral 3 14B
22.3
236
Claude Fable 5.1 (high with fallback)
21.9
237
Claude Opus 5 (xhigh)
21.6
238
Claude Opus 4.5
21.3
239
GPT-6 Astra (high)
20.9
240
Claude Opus 5 (high)
20.6
241
Claude Opus 5 (medium)
20.3
242
GPT-6 Astra (low)
19.9
243
Claude Opus 4.7 (max)
19.6
244
Grok 4.6 (low)
19.3
245
GPT-6 Astra (medium)
18.9
246
Qwen3.8-Flash-Next
18.6
247
Qwen3.8 27B (low)
18.3
248
GPT-6.1 Sol (low)
17.9
249
GLM-4.5V (non-reasoning)
17.6
250
Qwen3.8 27B (non-reasoning)
17.3
251
Kimi K3 (low)
16.9
252
Qwen3.8 27B (medium)
16.6
253
Claude 4.5 Sonnet
16.3
254
Kimi K2.6
15.9
255
MiMo-V2.6-Pro
15.6
256
DeepSeek V3 (Dec)
15.3
257
Claude Opus 4.5 (non-reasoning)
15.0
258
Claude Opus 4.7 (non-reasoning, high)
14.6
259
Claude Sonnet 4.6 (non-reasoning, high)
14.3
260
Kimi K2.6 (non-reasoning)
14.0
261
Qwen3.8 27B (xhigh)
13.6
262
Kimi K3 (max)
13.3
263
Gemma 4 31B (non-reasoning)
13.0
264
Mistral Medium 3
12.6
265
Claude Sonnet 4.6 (non-reasoning, low)
12.3
266
Qwen3 235B 2507
12.0
267
GLM-4.5V
11.6
268
Qwen3 30B A3B 2507 (non-reasoning)
11.3
269
GLM-4.7
11.0
270
Qwen3.8 Max (0902)
10.6
271
Claude 4.5 Sonnet (non-reasoning)
10.3
272
Seed-OSS-36B-Instruct
10.0
273
Gemma 4 E4B
9.6
274
GLM-4.7 (non-reasoning)
9.3
275
Qwen3.8 2.4T A95B
9.0
276
Claude Opus 4.6 (max)
8.6
277
Qwen3.8 Max
8.3
278
MiMo-V2.5-Pro
8.0
279
DeepSeek V3.1 Terminus (non-reasoning)
7.6
280
Claude Opus 4.6 (non-reasoning, high)
7.3
281
Mistral Small 3.2
7.0
282
DeepSeek V4 Flash (high)
6.6
283
DeepSeek V4 Flash (max)
6.3
284
MiMo-V2.5-Pro (non-reasoning)
6.0
285
GLM-4.6
5.6
286
Gemma 4 31BFREE
5.3
287
MiMo-V2.5
5.0
288
Mistral Small 3.1
4.7
289
GLM-4.6 (non-reasoning)
4.3
290
GPT-4 Turbo
4.0
291
Qwen2.5 72B
3.7
292
Gemma 4 E4B (non-reasoning)
3.3
293
DeepSeek R1 0528
3.0
294
Qwen3 32B
2.7
295
Qwen3 VL 30B A3B (指令)
2.3
296
Qwen3 VL 235B A22B (指令)
2.0
297
Qwen3.5 4B
1.7
298
Mixtral 8x7B
1.3
299
Qwen3 14B (non-reasoning)
1.0
300
Qwen3 14B
0.7
301
Qwen3.5 4B (non-reasoning)
0.3