Benchmark report
Is the OpenAI Decisions API slow?
Response times from 6 cities
Slow from some locations
A full yes/no answer typically came back in 120 ms from San Francisco and 312 ms from Mumbai, but the slower requests took up to 1.2 seconds.
- Fastest
- 120 ms
- San Francisco
- Slowest
- 312 ms
- Mumbai
- Successful requests
- 100%
- 6 locations
Response time around the world
San Francisco120 ms
Montreal156 ms
Tokyo202 ms
Amsterdam213 ms
Singapore299 ms
Mumbai312 ms
Why are the slower requests so much slower?
In Mumbai the typical answer took 312 ms, but the slowest 5% of requests took longer than 1,235 ms. Our half-hourly status checks show where that comes from: the first request after a quiet spell took nearly two seconds in Mumbai and about a second in Amsterdam, while the requests right after it took 150 to 400 ms. Connecting and the secure handshake took under 25 ms in every city, so the wait is on OpenAI's side.
Click the location to see each stage.
Mumbai324 ms
Mumbai- Finding the server
- 2 ms
- Reaching the server
- 3 ms
- Setting up security
- 6 ms
- Waiting for the server: most of the time
- 315 ms
- Receiving the response
- 0 ms
- Total
- 324 ms
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
Compare all locationsHide the comparison
Click a city to see its stage-by-stage breakdown.
Amsterdam409 ms
Amsterdam- Finding the server
- 1 ms
- Reaching the server
- 9 ms
- Setting up security
- 12 ms
- Waiting for the server: most of the time
- 388 ms
- Receiving the response
- 0 ms
- Total
- 409 ms
San Francisco124 ms
San Francisco- Finding the server
- 12 ms
- Reaching the server
- 3 ms
- Setting up security
- 6 ms
- Waiting for the server: most of the time
- 114 ms
- Receiving the response
- 1 ms
- Total
- 124 ms
Montreal151 ms
Montreal- Finding the server
- 1 ms
- Reaching the server
- 3 ms
- Setting up security
- 4 ms
- Waiting for the server: most of the time
- 143 ms
- Receiving the response
- 1 ms
- Total
- 151 ms
Singapore259 ms
Singapore- Finding the server
- 2 ms
- Reaching the server
- 3 ms
- Setting up security
- 4 ms
- Waiting for the server: most of the time
- 252 ms
- Receiving the response
- 0 ms
- Total
- 259 ms
Tokyo210 ms
Tokyo- Finding the server
- 1 ms
- Reaching the server
- 2 ms
- Setting up security
- 7 ms
- Waiting for the server: most of the time
- 201 ms
- Receiving the response
- 0 ms
- Total
- 210 ms
Mumbai324 ms
Mumbai- Finding the server
- 2 ms
- Reaching the server
- 3 ms
- Setting up security
- 6 ms
- Waiting for the server: most of the time
- 315 ms
- Receiving the response
- 0 ms
- Total
- 324 ms
The phases come from the daily run of 20 October 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
- API incidents are posted there. A slow answer during one is not your network.
- Expect a slow first request
- After a quiet period the first request took one to two seconds outside the US in our checks; the next ones were much faster.
- This is a full answer, not a first token
- The Decisions API does not stream, so we time the whole JSON response. The ChatGPT page times the first visible word of a streamed reply, which is a different measurement.
Response time over the last 9 days
A dot marks a day well above that location’s usual level.
Show the numbers
| Day | Amsterdam | San Francisco | Montreal | Singapore | Tokyo | Mumbai |
|---|---|---|---|---|---|---|
| 11 October 2026 | 133 ms | 98 ms | 151 ms | 270 ms | 191 ms | 126 ms |
| 12 October 2026 | 233 ms | 120 ms | 161 ms | 278 ms | 203 ms | 140 ms |
| 13 October 2026 | 148 ms | 111 ms | 109 ms | 299 ms | 200 ms | 307 ms |
| 14 October 2026 | 770 ms ▲ | 119 ms | 117 ms | 270 ms | 201 ms | 440 ms |
| 15 October 2026 | 127 ms | 107 ms | 165 ms | 299 ms | 187 ms | 326 ms |
| 16 October 2026 | 254 ms | 275 ms ▲ | 153 ms | 337 ms | 199 ms | 369 ms |
| 17 October 2026 | 140 ms | 148 ms | 253 ms | 324 ms | 209 ms | 112 ms |
| 18 October 2026 | 134 ms | 111 ms | 156 ms | 332 ms | 208 ms | 339 ms |
| 19 October 2026 | 255 ms | 120 ms | 155 ms | 271 ms | 194 ms | 312 ms |
| 20 October 2026 | 409 ms ▲ | 124 ms | 151 ms | 259 ms | 210 ms | 324 ms |
About this measurement
POST api.openai.com/v1/decisions · 50 requests per location over 10 daily runs · 11 October 2026 to 20 October 2026
What we tested · AI decision. Whole JSON response from gpt-6-luna to one yes/no question on a two-sentence support message: POST api.openai.com/v1/decisions, public beta, no streaming.
How LatencyRadar measures response time →Also measured: ChatGPT API →
Technical details
Doesn’t measure: Time to first token, Responses or Chat Completions, choice or score questions, images, longer inputs, or a fixed model: gpt-6-luna can change without notice.
One OpenAI surface: POST api.openai.com/v1/decisions with gpt-6-luna, a two-sentence support message and one yes/no question, paid key. The endpoint is a public beta and returns its answer as one JSON document, so the number here is the time to the complete answer on a new connection. The ChatGPT page measures OpenAI's chat API, which streams.
Between 11 and 20 October 2026 the typical time was 120 ms in San Francisco, 156 ms in Montreal, 202 ms in Tokyo, 213 ms in Amsterdam, 299 ms in Singapore and 312 ms in Mumbai. The slower requests varied much more: 95% finished within 310 ms in Tokyo, against 953 ms in Amsterdam and 1,235 ms in Mumbai. Tokyo was the steadiest city, with daily medians between 187 and 210 ms; Amsterdam's swung between 127 and 770 ms.
| Test location | Typical response time | Slower response time | Requests |
|---|---|---|---|
| Amsterdam | 213 ms | 953 ms | 50 |
| San Francisco | 120 ms | 344 ms | 50 |
| Montreal | 156 ms | 574 ms | 50 |
| Singapore | 299 ms | 840 ms | 50 |
| Tokyo | 202 ms | 310 ms | 50 |
| Mumbai | 312 ms | 1235 ms | 50 |
Typical: half of the requests finished within this time (technical: p50). Slower: 95% of requests finished within this time (technical: p95). Statistics
- Request
- POST api.openai.com/v1/decisions
- Measured from
- Amsterdam · San Francisco · Montreal · Singapore · Tokyo · Mumbai
- Requests
- Daily measurements from 11 October 2026 to 20 October 2026. 6 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo, Mumbai. · 11 October 2026 to 20 October 2026
- Model
- gpt-6-luna, the only model on the endpoint; no dated version, so results can change without notice · no streaming · whole JSON response timed
- Request
- One predicate question ("The customer is explicitly asking for a refund") about a two-sentence support message asking for a duplicate charge to be refunded. Identical in every city and on every day
- Timings taken
- DNS, connect, TLS, waiting for the server, download
- Report coverage
- 6 of 6 test locations included: Amsterdam, San Francisco, Montreal, Singapore, Tokyo, Mumbai.
- Daily requests in each location
- Amsterdam: 50 · San Francisco: 50 · Montreal: 50 · Singapore: 50 · Tokyo: 50 · Mumbai: 50
- 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.
OpenAI Decisions API
https://api.openai.com/v1/decisions
Independent measurement by LatencyRadar. Not affiliated with OpenAI. · How we measure (v1.3)
How fast is your API compared to OpenAI's Decisions API?
Run a free speed test from multiple cities and find out where your users are waiting. No setup, no account required.
No account required · Takes about 30 seconds.
