Live status: last 24 hours →

Benchmark report

ChatGPT API thumbnail

Is the ChatGPT API slow?

GPT-5.6 time to first token from 6 cities, measured daily

Uneven around the world

First visible token arrived in 748 ms from Montreal and 1326 ms from Mumbai. Some test locations waited noticeably longer, which suggests a more distant server for them.

Fastest
748 ms
Montreal
Slowest
1326 ms
Mumbai
Successful requests
100%
6 locations

Response time around the world

Amsterdam 1025ms typicalSan Francisco 928ms typicalMontreal 748ms typicalSingapore 977ms typicalTokyo 1003ms typicalMumbai 1326ms typical
  • CanadaMontreal
    748 ms
  • United StatesSan Francisco
    928 ms
  • SingaporeSingapore
    977 ms
  • JapanTokyo
    1003 ms
  • NetherlandsAmsterdam
    1025 ms
  • IndiaMumbai
    1326 ms

Almost all of the second is the model

OpenAI holds the response until the first token is ready: headers and first visible token arrived within 23 to 38 ms of each other in every city. Nearly all of the wait is queueing plus the model's low-effort reasoning before the stream opens. Connecting and the secure handshake stayed under 25 ms in every city, Mumbai included, which is Cloudflare's edge. What remained in the controlled run of 27 September was 0.6 to 1.2 seconds of OpenAI-side time, shortest in Montreal at 635 ms and longest in Mumbai at 1,197 ms.

Compare all locations

Click a city to see its stage-by-stage breakdown.

ConnectTLSServer waitDownload
NetherlandsAmsterdam
912 ms
Finding the server
2 ms
Reaching the server
10 ms
Setting up security
14 ms
Waiting for the server: most of the time
855 ms
Receiving the response
33 ms
Total
912 ms
Response opened
869 ms
First visible word
902 ms
Answer complete
912 ms
United StatesSan Francisco
879 ms
Finding the server
12 ms
Reaching the server
4 ms
Setting up security
6 ms
Waiting for the server: most of the time
830 ms
Receiving the response
39 ms
Total
879 ms
Response opened
841 ms
First visible word
871 ms
Answer complete
879 ms
CanadaMontreal
693 ms
Finding the server
1 ms
Reaching the server
3 ms
Setting up security
5 ms
Waiting for the server: most of the time
635 ms
Receiving the response
50 ms
Total
693 ms
Response opened
645 ms
First visible word
681 ms
Answer complete
693 ms
SingaporeSingapore
935 ms
Finding the server
13 ms
Reaching the server
4 ms
Setting up security
9 ms
Waiting for the server: most of the time
887 ms
Receiving the response
35 ms
Total
935 ms
Response opened
900 ms
First visible word
923 ms
Answer complete
935 ms
JapanTokyo
852 ms
Finding the server
2 ms
Reaching the server
3 ms
Setting up security
6 ms
Waiting for the server: most of the time
810 ms
Receiving the response
33 ms
Total
852 ms
Response opened
806 ms
First visible word
844 ms
Answer complete
852 ms
IndiaMumbai
1245 ms
Finding the server
6 ms
Reaching the server
3 ms
Setting up security
7 ms
Waiting for the server: most of the time
1197 ms
Receiving the response
38 ms
Total
1245 ms
Response opened
1207 ms
First visible word
1238 ms
Answer complete
1245 ms

The phases come from the controlled run of 27 September 2026. Typical and slower times pool the daily requests. 6 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo, Mumbai.

Finding the server is measured once per location, so it sits outside these bars. Open a city to see it.

Bars show the typical request, so the parts add up to its total. Why

OpenAI still feels slow?

Check OpenAI’s status page
Elevated latency and error rates are posted there. A slow first token during an incident is not your region.
A closer region will not help much
The wait is queueing and reasoning on OpenAI's side. Five of our six cities sat within about 40% of each other on typical first token. Mumbai was about 80% slower than Montreal, yet its connection to the API was as quick as anyone's, so the difference is behind the edge. Network evidence does not tell us where OpenAI runs the model.

Response time over the last 29 days

Daily median of 5 requests per location, 11 September 2026 to 10 October 2026.
025005000 ms11 Sept10 Oct
Show the numbers
DayAmsterdamSan FranciscoMontrealSingaporeTokyoMumbai
11 September 20261429 ms1209 ms1041 ms1698 ms936 ms1615 ms
12 September 20261203 ms1160 ms996 ms1551 ms762 ms1088 ms
13 September 20261607 ms701 ms959 ms1562 ms893 ms1463 ms
14 September 20261170 ms1291 ms1515 ms1985 ms922 ms1528 ms
15 September 20261095 ms1365 ms1353 ms1135 ms1346 ms1325 ms
16 September 20261293 ms1570 ms1043 ms961 ms1121 ms1689 ms
17 September 2026951 ms888 ms1001 ms1222 ms999 ms1112 ms
18 September 20261653 ms1133 ms778 ms1844 ms911 ms2433 ms
19 September 20261063 ms1153 ms764 ms1114 ms1038 ms1704 ms
20 September 20261313 ms910 ms604 ms1083 ms784 ms1168 ms
21 September 2026906 ms1095 ms827 ms930 ms829 ms1244 ms
22 September 20261529 ms959 ms769 ms965 ms898 ms1658 ms
23 September 2026909 ms1075 ms782 ms1668 ms1308 ms1526 ms
24 September 2026823 ms1004 ms1035 ms1253 ms988 ms1439 ms
25 September 20261249 ms901 ms1011 ms1210 ms960 ms1851 ms
26 September 2026763 ms715 ms580 ms725 ms732 ms893 ms
27 September 2026876 ms900 ms722 ms892 ms1120 ms1650 ms
28 September 2026948 ms757 ms704 ms777 ms873 ms1297 ms
29 September 20261192 ms982 ms711 ms842 ms1177 ms1298 ms
30 September 20261206 ms928 ms787 ms1694 ms922 ms1387 ms
1 October 20261236 ms1124 ms890 ms1016 ms1347 ms1300 ms
2 October 20261062 ms964 ms687 ms907 ms1003 ms964 ms
3 October 2026853 ms643 ms651 ms820 ms730 ms1028 ms
4 October 2026983 ms796 ms701 ms1090 ms1024 ms1082 ms
5 October 20261055 ms771 ms776 ms866 ms900 ms1543 ms
6 October 20261174 ms942 ms748 ms1232 ms1039 ms1389 ms
7 October 20261063 ms1021 ms975 ms948 ms982 ms1498 ms
9 October 2026991 ms1326 ms803 ms1452 ms1004 ms1649 ms
10 October 2026829 ms978 ms649 ms927 ms1074 ms1235 ms

About this measurement

POST api.openai.com/v1/chat/completions · 65 requests per location over 13 daily runs · 27 September 2026 to 10 October 2026

What we tested · AI response. Time to first visible token from gpt-5.6-sol on a one-word prompt: streaming, reasoning_effort low, 256 max tokens, paid key.

How LatencyRadar measures response time →

Technical details

Doesn’t measure: The ChatGPT app, tokens per second, long prompts, other models, tool calls or the Batch API.

The catalog contains two OpenAI requests. The primary is time to first visible token from gpt-5.6-sol at api.openai.com/v1/chat/completions: a one-word prompt, streaming, reasoning_effort low, 256 max tokens, paid key. It stands for the wait before the first word a user sees when calling the flagship model with light reasoning. The second is GET /v1/models with the same key, a small JSON list, which stands for the network and authentication path alone; no model runs, so it separates the front door from inference. Its results stay separate because a model-list response does not measure inference.

The published measurements cover 27 September to 10 October 2026. Typical first-token times range from 748 ms in Montreal to 1,326 ms in Mumbai, with the other four cities between 928 and 1,025 ms. Mumbai is the one city that stands apart: it was the slowest of the six on nine of the 13 days and never the fastest, and its daily median ranged from 964 to 1,650 ms. Its connection and handshake to Cloudflare's Mumbai edge took 10 ms in the controlled run, so the extra wait is on OpenAI's side of the edge, not on the way to it. Inside any one city the day-to-day movement is still large: Singapore's daily median alone ranged from 777 to 1,694 ms.

Test locationTypical response timeSlower response timeRequests
Amsterdam1025 ms1612 ms65
San Francisco928 ms1517 ms65
Montreal748 ms1424 ms65
Singapore977 ms1963 ms65
Tokyo1003 ms1993 ms65
Mumbai1326 ms2316 ms65

Typical: half of the requests finished within this time (technical: p50 of time to first visible token). Slower: 95% of requests finished within this time (technical: p95). Statistics

Request
Authenticated streaming POST to api.openai.com/v1/chat/completions
Measured from
Amsterdam · San Francisco · Montreal · Singapore · Tokyo · Mumbai
Requests
Daily measurements from 27 September 2026 to 10 October 2026. 6 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo, Mumbai. · 27 September 2026 to 10 October 2026
Model
gpt-5.6-sol · reasoning effort "low" · max_completion_tokens 256 · default temperature
Prompt
"Say 'ok' and nothing else." Identical across every city and every LLM API we benchmark
Network floor
Measured separately by an unauthenticated probe of the same hostname on 22 August 2026, from the five original cities
Comparison figures
Claude (claude-opus-5, effort low) and Grok (grok-4.6, reasoning_effort low) come from the same daily collection over the same days
Timings taken
Parsed from the token stream: stream open, first visible token, completion, plus DNS, connect and TLS
Report coverage
6 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo, Mumbai.
Coverage changes
Mumbai (measured since 7 September 2026) joined this report on 21 September 2026. Earlier daily bands keep their previous location set.
Daily requests in each location
Amsterdam: 65 · San Francisco: 65 · Montreal: 65 · Singapore: 65 · Tokyo: 65 · Mumbai: 65
Typical response time
Median of the eligible daily requests in each included location.
Slower response time
The 95th percentile of the same eligible daily requests in each included location.

ChatGPT API

https://api.openai.com/v1/chat/completions

Independent measurement by LatencyRadar. Not affiliated with OpenAI. · How we measure (v1.3)

How fast is your API compared to OpenAI's?

Run a free speed test from multiple cities and find out where your users are waiting. No setup, no account required.

Test my API

No account required · Takes about 30 seconds.

More benchmarks

All benchmarks →