Fastest AI APIs from Mumbai
How long five AI APIs take to start answering from a server in Mumbai, India, measured every day. Useful if your app runs there or serves users in India.
Mistral is the fastest AI API we measure from Mumbai: 326 ms to the first word, on its default settings.
Of the APIs that use low reasoning effort, Claude is fastest at 1,251 ms, ahead of ChatGPT (1,326 ms) and Grok (2,641 ms).
From Mumbai, ChatGPT takes 578 ms longer to start answering than from Montreal, its fastest city.
Which AI API answers fastest from Mumbai?
Typical time until the first word of the reply arrives, measured from Mumbai, 27 September to 10 October 2026.
MistralMistral AI · Provider default settings · Live status
326 ms typical
p95 TTFT: 748 ms
DeepSeekDeepSeek · Reasoning turned off · Live status
589 ms typical
p95 TTFT: 834 ms
ClaudeAnthropic · Low reasoning effort · Live status
1,251 ms typical
p95 TTFT: 1,665 ms
ChatGPTOpenAI · Low reasoning effort · Live status
1,326 ms typical
p95 TTFT: 2,316 ms
GrokxAI · Low reasoning effort
2,641 ms typical
p95 TTFT: 5,010 ms
Mistral runs on its default settings, DeepSeek runs with reasoning turned off, and Claude, ChatGPT and Grok use low reasoning effort. Typical is the middle request of the period; p95 means 95% of requests got their first word within that time.
Technical details
| API | Model | Requests in the period | Failed |
|---|---|---|---|
| Mistral | mistral-medium-latest | 65 | 0 of 65 |
| DeepSeek | deepseek-v4-flash | 65 | 0 of 65 |
| Claude | claude-opus-5 | 65 | 0 of 65 |
| ChatGPT | gpt-5.6-sol | 65 | 0 of 65 |
| Grok | grok-4.6 | 65 | 0 of 65 |
Compared to the fastest city
How much longer each API takes to start answering from Mumbai than from its own fastest test city.
How steady each API was from Mumbai
Each API's daily time from Mumbai, 11 September to 10 October 2026: one point per day, the middle of that day's five requests. Each chart has its own scale.
MistralSteady
9 days in 10 no more than 12% above its usual 335 ms
No failed requests · one day without a run
DeepSeekVaries
9 days in 10 no more than 22% above its usual 613 ms
No failed requests · one day without a run
ClaudeVaries
9 days in 10 no more than 17% above its usual 1,450 ms
No failed requests · one day without a run
ChatGPTVaries
9 days in 10 no more than 22% above its usual 1,389 ms
No failed requests · one day without a run
GrokVaries
9 days in 10 no more than 21% above its usual 2,741 ms
No failed requests · one day without a run
What to do with this
- Base timeouts on the slower figure: from Mumbai, 1 in 20 requests to Grok waited more than 5,010 ms for the first word.
- For predictable times, Mistral was the steadiest from Mumbai: on 9 days in 10 it stayed within 12% of its usual time. ChatGPT swung the most: on 1 day in 10 it was 22% or more above its usual time.
- If your backend doesn't have to run in Mumbai: ChatGPT took 748 ms from Montreal, against 1,326 ms here. Mistral took 214 ms from Amsterdam, against 326 ms here. Your users' trip to that server comes on top, so moving only helps when the API call is the slow part.
Good to know
- This is one server in a datacenter in Mumbai, measured once a day.
- Every API gets the same one-line prompt, but the reasoning settings differ (named on each row), so compare APIs with the same setting.
- Time to first token stops at the first visible word. It doesn't measure how fast the rest of the reply arrives or how good it is.
How fast is your API from Mumbai?
Run a free test from Mumbai and the other test cities, and see where the time goes.
Free, no account needed.