Which AI API has the lowest latency?

As of November 08, 2026, fireworks has the lowest measured edge latency at 16 ms p50 (time-to-first-byte, measured from Europe (Germany)). Rankings vary by region.

How is AI API latency measured here?

Distributed probes in multiple regions measure DNS, TCP, TLS and time-to-first-byte to each provider's API endpoint directly — not through a gateway — on a repeating schedule. Rankings are edge latency (TTFB); inference time-to-first-token is measured separately where we hold an API key. See the methodology page.

Is this latency data free to use?

Yes. All rankings are free and licensed CC-BY-4.0, with a JSON API, per-page markdown, and an MCP endpoint for programmatic and AI-agent use.

AI API latency

Compare endpoint response times and availability across regions.

Download JSON

Overview Latency Uptime IncidentsLast probe · 08 Nov, 04:40 UTC

Asia (Tokyo)fastest · Fireworks AI19 ms
Europe (Germany)fastest · Fireworks AI16 ms
South America (São Paulo)fastest · OpenRouter59 ms
US (Central)fastest · Fireworks AI29 ms
Asia (Tokyo)fastest · Fireworks AI19 ms
Europe (Germany)fastest · Fireworks AI16 ms
South America (São Paulo)fastest · OpenRouter59 ms
US (Central)fastest · Fireworks AI29 ms

Asia (Tokyo)fastest · Fireworks AI19 ms
Europe (Germany)fastest · Fireworks AI16 ms
South America (São Paulo)fastest · OpenRouter59 ms
US (Central)fastest · Fireworks AI29 ms
Asia (Tokyo)fastest · Fireworks AI19 ms
Europe (Germany)fastest · Fireworks AI16 ms
South America (São Paulo)fastest · OpenRouter59 ms
US (Central)fastest · Fireworks AI29 ms

Fastest typical response

16ms

Fireworks AI · Europe (Germany)

Endpoint availability

99.9%

50,423 network probes · last 24h

Providers

API endpoints monitored

Regions

Independent probe locations

Response time over 24 hours

Typical response (p50 TTFB) · Europe (Germany)

Lower is faster

OpenAIAnthropicGoogleUTC · lower is faster

View hourly values

Hourly median (p50), in milliseconds. Times are UTC. A dash means no measurement.

Hourly endpoint response times
Date · UTC	OpenAI	Anthropic	Google
11-08 04:00	322	198	86
11-08 03:00	248	188	93
11-08 02:00	209	192	84
11-08 01:00	312	189	92
11-08 00:00	319	198	27
11-07 23:00	235	197	93
11-07 22:00	280	200	93
11-07 21:00	297	200	97
11-07 20:00	206	198	96
11-07 19:00	221	199	98
11-07 18:00	298	200	95
11-07 17:00	328	201	85
11-07 16:00	323	197	91
11-07 15:00	212	197	89
11-07 14:00	219	199	97
11-07 13:00	211	197	97
11-07 12:00	215	196	97
11-07 11:00	317	200	96
11-07 10:00	207	195	84
11-07 09:00	315	199	95
11-07 08:00	213	198	88
11-07 07:00	323	199	88
11-07 06:00	209	197	88
11-07 05:00	205	197	28

Looking for model response time? View time to first token (TTFT)

Speed index

Relative endpoint speed · 4 measured regions

Higher is faster

01Fireworks AI81/100 02OpenRouter51/100 03Google41/100 04baseten31/100 05Cohere27/100

100 matches the fastest endpoint in every measured region. 50 means twice the latency in a region; scores are averaged across regions.

Full rankings & scoring method

Provider rankings

Choose the region closest to your application’s server · last 24 hours

Endpoint latency

How to read these metrics

Typical response · p50Half of successful timed probes were this fast or faster. Lower is better.

Slower responses · p9595% of successful timed probes were this fast or faster. This shows how latency behaves on slower requests.

AvailabilityThe share of successful network probes. Higher is better. Samples show how many probes were made.

TTFB measures the network and API endpoint, before a model generates an answer. It does not measure answer quality or generation speed. Compare model TTFT or read the measurement method. A dash means no timing data.

Asia (Tokyo)

Rank	Provider	Typical responsep50 TTFB · ms	Slower responsesp95 TTFB · ms	AvailabilitySuccessful probes	SamplesLast 24 hours
01	Fireworks AI	19 ms	61 ms	100%	288
02	SambaNova	31 ms	454 ms	100%	288
03	OpenRouter	53 ms	115 ms	100%	288
04	upstage	59 ms	78 ms	100%	288
05	Google	62 ms	120 ms	100%	288
06	glm	94 ms	140 ms	100%	288
07	meta-llama	117 ms	295 ms	100%	288
08	kimi	130 ms	213 ms	100%	288
09	ernie	138 ms	166 ms	100%	288
10	baichuan	142 ms	564 ms	99.3%	288
11	TypeSafe	144 ms	151 ms	100%	288
12	novita	150 ms	216 ms	100%	288
13	baseten	160 ms	183 ms	100%	288
14	Together AI	161 ms	238 ms	100%	288
15	Groq	163 ms	234 ms	100%	288
16	qwen	166 ms	227 ms	100%	288
17	xAI	171 ms	252 ms	100%	288
18	Cohere	174 ms	223 ms	100%	288
19	replicate	182 ms	454 ms	100%	288
20	DeepSeek	182 ms	251 ms	100%	288
21	Cerebras	199 ms	256 ms	100%	288
22	OpenAI	213 ms	294 ms	100%	288
23	Perplexity	223 ms	408 ms	100%	288
24	minimax	224 ms	258 ms	100%	288
25	Anthropic	230 ms	295 ms	100%	288
26	iflytek	240 ms	911 ms	99.7%	288
27	reka	241 ms	526 ms	100%	288
28	writer	244 ms	278 ms	100%	288
29	sarvam	251 ms	258 ms	100%	288
30	Tencent TokenHub	256 ms	297 ms	100%	126
31	friendli	280 ms	609 ms	100%	288
32	ai21	313 ms	421 ms	100%	288
33	inference-net	316 ms	616 ms	100%	288
34	Mistral	319 ms	386 ms	100%	288
35	stepfun	326 ms	849 ms	100%	288
36	siliconflow	374 ms	403 ms	100%	288
37	hyperbolic	403 ms	528 ms	100%	288
38	featherless	426 ms	762 ms	100%	288
39	sensenova	440 ms	1,161 ms	98.6%	288
40	deepinfra	448 ms	926 ms	100%	288
41	nebius	466 ms	502 ms	100%	288
42	doubao	495 ms	1,444 ms	99.0%	288
43	nscale	498 ms	543 ms	100%	288
44	aleph-alpha	544 ms	573 ms	100%	288

44 providersRegion report

Europe (Germany)

Rank	Provider	Typical responsep50 TTFB · ms	Slower responsesp95 TTFB · ms	AvailabilitySuccessful probes	SamplesLast 24 hours
01	Fireworks AI	16 ms	93 ms	100%	298
02	nebius	33 ms	98 ms	100%	298
03	aleph-alpha	62 ms	96 ms	100%	298
04	nscale	69 ms	74 ms	100%	298
05	meta-llama	88 ms	123 ms	100%	298
06	Google	92 ms	103 ms	100%	298
07	OpenRouter	98 ms	105 ms	100%	298
08	Mistral	118 ms	200 ms	100%	298
09	baseten	125 ms	131 ms	100%	298
10	Cohere	133 ms	200 ms	100%	298
11	TypeSafe	156 ms	165 ms	100%	298
12	replicate	170 ms	325 ms	100%	298
13	glm	183 ms	204 ms	100%	298
14	baichuan	183 ms	625 ms	99.7%	297
15	reka	185 ms	211 ms	100%	298
16	Perplexity	190 ms	208 ms	100%	298
17	Cerebras	194 ms	207 ms	100%	298
18	Anthropic	197 ms	206 ms	100%	298
19	Groq	213 ms	224 ms	100%	298
20	OpenAI	219 ms	349 ms	100%	298
21	writer	224 ms	289 ms	100%	298
22	inference-net	225 ms	610 ms	100%	298
23	novita	230 ms	248 ms	100%	298
24	Together AI	232 ms	294 ms	100%	298
25	siliconflow	248 ms	250 ms	100%	298
26	friendli	257 ms	436 ms	100%	298
27	sarvam	260 ms	269 ms	100%	298
28	iflytek	283 ms	754 ms	100%	298
29	ernie	287 ms	317 ms	100%	297
30	DeepSeek	290 ms	306 ms	100%	298
31	xAI	304 ms	323 ms	100%	298
32	kimi	310 ms	393 ms	100%	298
33	minimax	313 ms	344 ms	100%	298
34	ai21	317 ms	448 ms	100%	298
35	SambaNova	320 ms	330 ms	100%	298
36	doubao	351 ms	845 ms	99.7%	298
37	qwen	361 ms	444 ms	99.7%	298
38	sensenova	388 ms	1,022 ms	99.7%	297
39	stepfun	428 ms	1,143 ms	100%	297
40	Tencent TokenHub	466 ms	641 ms	100%	126
41	deepinfra	507 ms	936 ms	100%	298
42	hyperbolic	568 ms	756 ms	100%	298
43	upstage	586 ms	596 ms	100%	297
44	featherless	587 ms	1,276 ms	100%	298

44 providersRegion report

South America (São Paulo)

Rank	Provider	Typical responsep50 TTFB · ms	Slower responsesp95 TTFB · ms	AvailabilitySuccessful probes	SamplesLast 24 hours
01	OpenRouter	59 ms	100 ms	100%	287
02	Google	139 ms	188 ms	100%	290
03	baseten	154 ms	170 ms	100%	288
04	Cohere	164 ms	199 ms	100%	287
05	TypeSafe	174 ms	196 ms	100%	288
06	Cerebras	188 ms	240 ms	100%	288
07	Anthropic	193 ms	248 ms	100%	290
08	Perplexity	203 ms	430 ms	100%	288
09	reka	204 ms	278 ms	100%	288
10	writer	211 ms	275 ms	100%	287
11	inference-net	218 ms	456 ms	100%	287
12	xAI	221 ms	383 ms	100%	288
13	Groq	232 ms	288 ms	100%	288
14	OpenAI	233 ms	414 ms	100%	290
15	meta-llama	239 ms	339 ms	100%	288
16	novita	246 ms	286 ms	100%	287
17	Fireworks AI	253 ms	278 ms	100%	288
18	Together AI	261 ms	638 ms	100%	288
19	Mistral	268 ms	340 ms	100%	289
20	siliconflow	284 ms	288 ms	100%	288
21	friendli	300 ms	568 ms	100%	288
22	ai21	329 ms	458 ms	100%	288
23	kimi	339 ms	398 ms	100%	288
24	baichuan	359 ms	1,250 ms	100%	288
25	SambaNova	370 ms	388 ms	100%	288
26	nebius	376 ms	389 ms	100%	288
27	ernie	411 ms	450 ms	100%	288
28	replicate	424 ms	597 ms	100%	287
29	DeepSeek	435 ms	506 ms	100%	288
30	aleph-alpha	446 ms	476 ms	99.7%	288
31	nscale	451 ms	454 ms	100%	288
32	minimax	480 ms	506 ms	100%	288
33	deepinfra	530 ms	918 ms	100%	288
34	glm	541 ms	573 ms	100%	288
35	upstage	612 ms	700 ms	100%	288
36	sarvam	622 ms	634 ms	100%	288
37	sensenova	638 ms	1,583 ms	99.0%	288
38	iflytek	657 ms	1,568 ms	99.7%	288
39	stepfun	670 ms	1,668 ms	100%	288
40	featherless	679 ms	1,008 ms	100%	288
41	doubao	703 ms	1,606 ms	98.6%	288
42	qwen	704 ms	753 ms	100%	288
43	hyperbolic	784 ms	903 ms	100%	288
44	Tencent TokenHub	794 ms	828 ms	100%	126

44 providersRegion report

US (Central)

Rank	Provider	Typical responsep50 TTFB · ms	Slower responsesp95 TTFB · ms	AvailabilitySuccessful probes	SamplesLast 24 hours
01	Fireworks AI	29 ms	80 ms	100%	288
02	Google	40 ms	112 ms	100%	287
03	baseten	47 ms	58 ms	100%	288
04	OpenRouter	57 ms	122 ms	100%	288
05	Cohere	61 ms	131 ms	100%	288
06	meta-llama	68 ms	172 ms	100%	288
07	Cerebras	73 ms	147 ms	100%	288
08	TypeSafe	78 ms	92 ms	100%	288
09	siliconflow	82 ms	109 ms	100%	288
10	Perplexity	84 ms	136 ms	100%	288
11	replicate	93 ms	150 ms	100%	288
12	Anthropic	93 ms	161 ms	100%	287
13	novita	100 ms	164 ms	100%	288
14	reka	100 ms	185 ms	100%	288
15	xAI	110 ms	182 ms	100%	288
16	writer	112 ms	138 ms	100%	288
17	Together AI	124 ms	189 ms	100%	288
18	Groq	128 ms	200 ms	100%	288
19	OpenAI	137 ms	226 ms	100%	287
20	deepinfra	157 ms	608 ms	100%	288
21	SambaNova	167 ms	218 ms	100%	288
22	friendli	178 ms	362 ms	99.7%	288
23	nebius	180 ms	215 ms	100%	288
24	inference-net	187 ms	456 ms	100%	288
25	Mistral	196 ms	269 ms	100%	287
26	ai21	233 ms	382 ms	100%	288
27	nscale	236 ms	243 ms	100%	288
28	baichuan	240 ms	865 ms	100%	288
29	glm	252 ms	304 ms	100%	288
30	aleph-alpha	256 ms	313 ms	100%	288
31	hyperbolic	261 ms	345 ms	100%	288
32	kimi	266 ms	332 ms	100%	288
33	ernie	272 ms	283 ms	100%	288
34	featherless	282 ms	643 ms	100%	288
35	DeepSeek	314 ms	386 ms	100%	287
36	minimax	343 ms	365 ms	100%	288
37	upstage	354 ms	405 ms	100%	288
38	stepfun	395 ms	1,370 ms	100%	288
39	sensenova	398 ms	1,423 ms	98.3%	288
40	qwen	411 ms	458 ms	100%	288
41	iflytek	420 ms	1,127 ms	99.7%	288
42	doubao	459 ms	1,152 ms	98.6%	288
43	sarvam	500 ms	523 ms	100%	288
44	Tencent TokenHub	573 ms	737 ms	100%	126

44 providersRegion report

Is any AI API down right now?

2 confirmed outage(s) in the last 24 hours: doubao, sensenova

44 providers · probed every 5 minutes from 4 regions · last probe Nov 8, 04:40 UTC

Status pages currently reporting an open incident: OpenAI (notice “Elevated error rates with some U.S. business Mastercards”) — self-reported by the provider, checked Nov 8, 04:17 UTC. Our probes test the API host; errors limited to individual models are visible only to the provider.

Live status per provider: OpenAI API · Claude API (Anthropic) · Gemini API (Google) · Mistral API · DeepSeek API · xAI API · Groq API · Cerebras API · SambaNova API · Together AI API · Fireworks AI API · OpenRouter API · TypeSafe API

Endpoint availability · hourly probe results

Last 72 hours · UTC · newest time on the right · current hour is incomplete.

One row per provider. Each block shows the results of one hour’s endpoint probes, combined across regions. Adjacent hours with all successes or no data are joined.

all probes succeededsome failed (under half)half or more failedno probes recorded

Probe counts by provider

OpenAI: 2 failed / 3,457 endpoint probesAnthropic: 1 failed / 3,457 endpoint probesGoogle: 0 failed / 3,456 endpoint probesMistral: 0 failed / 3,455 endpoint probesDeepSeek: 0 failed / 3,454 endpoint probesxAI: 0 failed / 3,454 endpoint probesGroq: 0 failed / 3,454 endpoint probesCerebras: 0 failed / 3,454 endpoint probesSambaNova: 5 failed / 3,454 endpoint probesTogether AI: 1 failed / 3,454 endpoint probesFireworks AI: 0 failed / 3,454 endpoint probesOpenRouter: 0 failed / 3,454 endpoint probesTypeSafe: 0 failed / 3,454 endpoint probesdoubao: 28 failed / 3,454 endpoint probessensenova: 15 failed / 3,453 endpoint probes

Full incident log

Fastest by region

Lowest typical response (p50 TTFB) · last 24 hours

Methodology

Asia (Tokyo)Fireworks AI19ms Europe (Germany)Fireworks AI16ms South America (São Paulo)OpenRouter59ms US (Central)Fireworks AI29ms

Which AI API is fastest right now?

As of November 08, 2026, the fastest-responding AI inference API by edge latency (time-to-first-byte) is fireworks at 16 ms p50, measured from Europe (Germany) (100% uptime, n=298). Rankings differ by region.

Edge latency (TTFB) measures the network and API endpoint; inference TTFT is measured separately. Methodology

All 44 providers

Response time over 24 hours

Speed index

Provider rankings

Asia (Tokyo)

Europe (Germany)

South America (São Paulo)

US (Central)

Is any AI API down right now?

Fastest by region

Which AI API is fastest right now?

Provider comparisons