榜单

首字延迟榜近 24 小时

从发出请求到收到第一个 token 的耗时，取近 24 小时的线上实测值。越低越好。

数据136 条线路 · 13 家中转站同步于 2026/11/13 04:59:04，每 15 分钟更新

模型：GPT 5.6 TerraGPT 6.1 SolGemini 3.1 Pro PreviewGPT 5.5Claude Fable 5Claude Haiku 5.5GLM 5.3Claude Opus 5Qwen 3.8 MaxGrok 4.6Kimi K3GPT 6 SolGPT 5.6 SolClaude Sonnet 5DeepSeek V4.1 FlashGemini 3.7 FlashGPT 6 AstraClaude Fable 5.1DeepSeek V4 ProGemini 3.8 FlashClaude Opus 5.5DeepSeek V4 FlashGemini 3.5 FlashGemini 3.6 FlashGrok 4.5Grok 4.7GLM 5.3 Flash

排名	模型 @ 中转站	首字延迟	样本	可用率	会话成本	数据安全
1	GPT 5.6 Terra@ Modelflare	0.86 s	111	97.3%	¥14.9	4
2	GPT 6.1 Sol@ Modelflare	0.87 s	148	98.7%	¥0.44	4
3	Gemini 3.1 Pro Preview@ Modelflare	0.91 s	110	98.2%	¥4.46	4
4	GPT 5.5@ Modelflare	0.93 s	118	97.5%	¥37.2	4
5	Claude Fable 5@ Modelflare	0.93 s	153	98.7%	¥36.5	4
6	Claude Haiku 5.5@ Modelflare	0.94 s	200	98.0%	—	4
7	GLM 5.3@ Modelflare	0.96 s	197	98.5%	¥8.56	4
8	Claude Opus 5@ Modelflare	0.99 s	177	97.2%	¥2.19	4
9	GPT 5.5@ Modelflare	1.01 s	161	98.1%	¥1.12	4
10	GPT 5.5@ Modelflare	1.01 s	184	97.8%	¥5.95	4
11	GPT 5.6 Terra@ Modelflare	1.03 s	192	99.0%	¥0.89	4
12	GPT 5.6 Terra@ Modelflare	1.04 s	141	97.9%	¥0.45	4
13	Qwen 3.8 Max@ Modelflare	1.06 s	147	98.0%	¥20.4	4
14	Grok 4.6@ Modelflare	1.16 s	197	99.0%	¥0.28	4
15	Kimi K3@ Modelflare	1.16 s	198	98.0%	¥28.6	4
16	GPT 6 Sol@ Modelflare	1.21 s	132	97.7%	¥0.44	4
17	Claude Opus 5@ Modelflare	1.21 s	195	99.0%	¥18.3	4
18	GPT 5.6 Sol@ Modelflare	1.24 s	105	98.1%	¥2.23	4
19	Claude Sonnet 5@ Modelflare	1.29 s	143	97.2%	¥0.88	4
20	GPT 5.6 Sol@ Modelflare	1.31 s	102	97.1%	¥1.12	4
21	DeepSeek V4.1 Flash@ Modelflare	1.31 s	217	98.2%	¥2.59	4
22	Gemini 3.7 Flash@ Modelflare	1.32 s	130	98.5%	¥3.29	4
23	Claude Fable 5@ Modelflare	1.32 s	104	98.1%	¥4.38	4
24	GPT 6 Astra@ Modelflare	1.33 s	135	98.5%	¥2.19	4
25	GPT 5.6 Sol@ Modelflare	1.34 s	138	97.8%	¥37.2	4
26	Claude Fable 5.1@ Modelflare	1.34 s	170	98.2%	¥4.38	4
27	DeepSeek V4 Pro@ Modelflare	1.35 s	199	99.0%	¥11.4	4
28	Gemini 3.8 Flash@ Modelflare	1.40 s	139	98.6%	¥3.29	4
29	Claude Opus 5@ Modelflare	1.44 s	166	97.6%	¥23.0	4
30	GPT 5.6 Terra@ Modelflare	1.56 s	138	98.5%	¥2.38	4
31	Claude Opus 5.5@ Modelflare	1.57 s	146	97.3%	¥1.75	4
32	Claude Sonnet 5@ Modelflare	1.58 s	172	97.7%	¥9.20	4
33	DeepSeek V4 Flash@ Modelflare	1.72 s	145	97.2%	¥2.59	4
34	Claude Fable 5@ Modelflare	1.77 s	146	98.6%	¥46.0	4
35	Claude Sonnet 5@ Modelflare	1.77 s	130	97.7%	¥7.30	4
36	Gemini 3.5 Flash@ Modelflare	1.78 s	103	98.1%	¥3.35	4
37	Gemini 3.6 Flash@ Modelflare	1.79 s	146	98.6%	¥3.29	4
38	GPT 5.5@ Modelflare	1.82 s	164	98.2%	¥2.23	4
39	DeepSeek V4.1 Flash@ 智流 FluxLane	1.83 s	30	73.3%	—	4
40	Grok 4.5@ Modelflare	1.84 s	126	97.6%	¥0.28	4
41	GPT 5.6 Sol@ Modelflare	1.84 s	134	97.8%	¥5.95	4
42	DeepSeek V4 Pro@ 智流 FluxLane	2.27 s	31	100.0%	—	4
43	Grok 4.6@ 智流 FluxLane	2.57 s	31	80.7%	—	4
44	Grok 4.6@ 智流 FluxLane	2.57 s	31	80.7%	—	4
45	Grok 4.5@ 智流 FluxLane	4.79 s	31	87.1%	—	4
46	Grok 4.5@ 智流 FluxLane	4.79 s	31	87.1%	—	4
47	Grok 4.7@ Modelflare	5.11 s	176	98.3%	¥3.10	4
48	Grok 4.5@ Modelflare	5.20 s	215	97.2%	¥3.10	4
49	Claude Fable 5@ 智流 FluxLane	5.48 s	29	93.1%	—	4
50	Claude Fable 5@ 智流 FluxLane	5.48 s	29	93.1%	—	4
51	Claude Fable 5@ 智流 FluxLane	5.48 s	29	93.1%	—	4
52	Grok 4.6@ Modelflare	6.17 s	197	98.5%	¥3.10	4
53	Claude Opus 5@ 智流 FluxLane	6.57 s	30	90.0%	—	4
54	Claude Opus 5@ 智流 FluxLane	6.57 s	30	90.0%	—	4
55	Claude Opus 5@ 智流 FluxLane	6.57 s	30	90.0%	—	4
56	Claude Fable 5.1@ 智流 FluxLane	7.00 s	30	96.7%	—	4
57	GPT 5.6 Sol@ 智流 FluxLane	7.46 s	31	100.0%	—	4
58	GPT 6.1 Sol@ 智流 FluxLane	8.85 s	31	100.0%	—	4
59	Claude Opus 5.5@ 智流 FluxLane	9.05 s	30	93.3%	—	4
60	GPT 6 Astra@ 智流 FluxLane	9.76 s	31	100.0%	—	4
61	GPT 5.6 Terra@ 智流 FluxLane	12.41 s	30	96.7%	—	4
62	Claude Sonnet 5@ 智流 FluxLane	23.19 s	31	100.0%	—	4
63	Claude Sonnet 5@ 智流 FluxLane	23.19 s	31	100.0%	—	4
64	Claude Sonnet 5@ 智流 FluxLane	23.19 s	31	100.0%	—	4
65	Kimi K3@ CheapAI	3.02 s	31	77.4%	¥4.40	3
66	Claude Fable 5@ CheapAI	5.01 s	31	100.0%	—	3
67	GPT 5.6 Terra@ CheapAI	12.36 s	31	64.5%	¥0.27	3
68	Claude Opus 5@ CheapAI	16.57 s	31	74.2%	—	3
69	GPT 5.6 Sol@ CheapAI	18.77 s	31	80.7%	¥0.67	3
70	Claude Opus 5.5@ CheapAI	19.67 s	31	77.4%	—	3
71	Claude Fable 5@ 境Ai	1.95 s	18	5.6%	—	0
72	Claude Opus 5.5@ Apiporter	2.03 s	18	100.0%	—	0
73	Claude Fable 5.1@ Apiporter	2.17 s	18	100.0%	—	0
74	Claude Opus 5.5@ Zashmodel	2.38 s	30	60.0%	¥4.77	0
75	GPT 6.1 Sol@ Apiporter	2.45 s	18	100.0%	—	0
76	DeepSeek V4.1 Flash@ ZNB.ai	2.51 s	30	100.0%	—	0
77	Claude Sonnet 5@ IKunCode	2.88 s	18	100.0%	¥4.40	0
78	GPT 6 Astra@ Apiporter	2.88 s	18	100.0%	—	0
79	Grok 4.7@ Cloudbridge	3.31 s	18	100.0%	—	0
80	GPT 5.6 Sol@ 境Ai	3.32 s	18	100.0%	¥4.48	0
81	GPT 6 Astra@ IKunCode	3.49 s	18	100.0%	¥15.4	0
82	Claude Sonnet 5@ MuskAI	3.52 s	31	100.0%	—	0
83	DeepSeek V4.1 Flash@ MuskAI	3.63 s	31	100.0%	—	0
84	DeepSeek V4.1 Flash@ IKunCode	3.95 s	18	100.0%	—	0
85	Claude Opus 5.5@ CheapRouter	3.97 s	18	100.0%	¥4.40	0
86	GPT 6 Astra@ CheapRouter	4.04 s	18	100.0%	¥3.30	0
87	DeepSeek V4.1 Flash@ CheapRouter	4.17 s	18	100.0%	¥1.30	0
88	Kimi K3@ NexoToken	4.34 s	30	96.7%	—	0
89	Grok 4.6@ IKunCode	4.59 s	18	100.0%	¥1.27	0
90	GPT 6.1 Sol@ Zashmodel	4.62 s	30	100.0%	¥2.38	0
91	GPT 6.1 Sol@ ZNB.ai	4.67 s	30	90.0%	—	0
92	GPT 5.6 Sol@ NexoToken	4.74 s	30	90.0%	—	0
93	Grok 4.5@ IKunCode	4.78 s	18	100.0%	¥1.27	0
94	Grok 4.7@ IKunCode	4.88 s	18	100.0%	¥1.27	0
95	GPT 5.6 Sol@ MuskAI	5.11 s	31	96.8%	—	0
96	GPT 6 Astra@ ZNB.ai	5.20 s	30	96.7%	—	0
97	Kimi K3@ A6api	5.53 s	18	88.9%	¥2.04	0
98	DeepSeek V4.1 Flash@ NexoToken	6.03 s	29	89.7%	—	0
99	Claude Opus 5@ IKunCode	6.04 s	18	94.4%	¥11.0	0
100	GPT 6 Astra@ MuskAI	6.04 s	30	100.0%	—	0
101	GPT 5.6 Sol@ NexoToken	6.39 s	30	90.0%	—	0
102	DeepSeek V4 Flash@ A6api	6.41 s	18	94.4%	¥2.04	0
103	Claude Opus 5@ NexoToken	6.45 s	30	100.0%	—	0
104	DeepSeek V4.1 Flash@ Cloudbridge	6.48 s	18	94.4%	—	0
105	DeepSeek V4 Pro@ A6api	6.48 s	18	88.9%	¥2.04	0
106	Claude Sonnet 5@ NexoToken	6.55 s	30	96.7%	—	0
107	GPT 6.1 Sol@ NexoToken	6.70 s	30	90.0%	—	0
108	GLM 5.3@ NexoToken	7.02 s	30	90.0%	—	0
109	GPT 6.1 Sol@ CheapRouter	7.18 s	18	100.0%	¥0.66	0
110	Claude Fable 5@ NexoToken	7.25 s	31	80.7%	—	0
111	GPT 6 Astra@ NexoToken	8.02 s	30	100.0%	—	0
112	GPT 6 Sol@ A6api	8.05 s	18	88.9%	¥2.04	0
113	GPT 6 Astra@ Zashmodel	8.16 s	30	93.3%	¥11.9	0
114	GPT 5.6 Terra@ NexoToken	9.10 s	30	100.0%	—	0
115	Claude Sonnet 5@ A6api	9.30 s	18	77.8%	¥2.04	0
116	DeepSeek V4 Pro@ NexoToken	9.57 s	30	96.7%	—	0
117	GLM 5.3 Flash@ NexoToken	9.59 s	30	80.0%	—	0
118	GPT 5.6 Sol@ NexoToken	9.64 s	30	100.0%	—	0
119	GPT 6 Astra@ A6api	9.99 s	18	100.0%	¥2.04	0
120	GPT 6 Sol@ NexoToken	10.13 s	29	100.0%	—	0
121	Grok 4.7@ NexoToken	10.99 s	30	96.7%	—	0
122	GPT 5.6 Sol@ NexoToken	11.22 s	29	89.7%	—	0
123	GLM 5.3 Flash@ MuskAI	11.76 s	31	96.8%	—	0
124	Claude Opus 5@ A6api	11.82 s	18	83.3%	¥2.04	0
125	Claude Opus 5.5@ IKunCode	12.04 s	18	77.8%	¥8.80	0
126	Claude Opus 5.5@ Cloudbridge	12.08 s	18	61.1%	—	0
127	GPT 5.6 Terra@ NexoToken	12.37 s	30	100.0%	—	0
128	Grok 4.6@ NexoToken	12.86 s	30	93.3%	—	0
129	Grok 4.7@ MuskAI	12.91 s	31	100.0%	—	0
130	DeepSeek V4 Flash@ NexoToken	13.78 s	30	90.0%	—	0
131	Claude Fable 5.1@ Cloudbridge	14.16 s	18	77.8%	—	0
132	Claude Opus 5.5@ A6api	15.94 s	18	88.9%	¥2.04	0
133	Grok 4.5@ NexoToken	18.80 s	30	96.7%	—	0
134	Claude Fable 5@ A6api	19.34 s	18	83.3%	¥2.04	0
135	Claude Opus 5.5@ ZNB.ai	21.85 s	29	82.8%	—	0
136	Claude Fable 5.1@ A6api	24.58 s	18	88.9%	¥2.04	0

常见问题

首字延迟榜怎么计算？

从发出请求到收到第一个 token 的耗时，取近 24 小时的线上实测值。越低越好。

为什么同一家中转站出现多次？

每一行是一条“模型 @ 中转站”线路。同一家中转站的不同模型、不同分组表现可能差别很大，所以分开排名。

首字延迟榜 近 24 小时

常见问题

首字延迟榜近 24 小时