DEEPSEEK LATENCY BENCHMARK · Sept 2026

DeepSeek direct or via OpenRouter?

We tested both from six cities. The faster option depends heavily on where you call from.

What the data shows

  1. Finding 1

    Location matters more than the route

    Direct was faster from Singapore and Mumbai. OpenRouter was often faster from Montreal.

    • Singapore and Mumbai: direct 621–664 ms, routes 782–1,348 ms
    • Montreal: routes 422–692 ms, direct 732 ms
    See the chart by city →

  2. Finding 2

    DeepSeek's direct API was more consistent

    DeepSeek's API had much lower p95 latency and less variation between locations.

    • City spread: 36% direct, 89–219% via OpenRouter
    • Slow days: 0 direct, 1 via OpenRouter
    See the day-by-day chart →

  3. Finding 3

    Most of the wait isn't the network

    Connection setup was only 11–19 ms. Most latency came after reaching the provider.

    • Network: 11–19 ms
    • First token: 462–883 ms
    See where the time goes →

Time to first token, day by day

Daily median time to first token, 27 August 2026 to 22 September 2026. Each point is the median of the compared cities’ daily medians, 5 requests per city.
0200040006000 ms27 Aug22 SeptOpenRouter · Together, 18 September 2026: 5566 ms, every compared city well above its usual level.

A dot marks a day when every compared city was well above that provider’s usual level.

Show the numbers
DayDeepSeek directOpenRouter · CloudflareOpenRouter · TogetherOpenRouter · DigitalOcean
27 August 2026——722 ms949 ms
28 August 2026—646 ms661 ms498 ms
29 August 2026——602 ms756 ms
31 August 2026——580 ms650 ms
1 September 2026——936 ms913 ms
2 September 2026——1586 ms540 ms
3 September 2026—796 ms784 ms827 ms
4 September 2026—654 ms—772 ms
5 September 2026714 ms533 ms765 ms639 ms
6 September 2026493 ms599 ms—666 ms
7 September 2026617 ms—875 ms799 ms
8 September 2026538 ms626 ms1057 ms805 ms
9 September 2026434 ms820 ms759 ms988 ms
10 September 2026781 ms743 ms829 ms764 ms
11 September 2026611 ms729 ms697 ms705 ms
12 September 2026645 ms873 ms859 ms736 ms
13 September 2026676 ms650 ms689 ms843 ms
14 September 2026802 ms748 ms757 ms792 ms
15 September 2026902 ms1274 ms687 ms752 ms
16 September 2026733 ms714 ms—695 ms
17 September 2026594 ms745 ms—861 ms
18 September 2026732 ms727 ms5566 ms ▲806 ms
19 September 2026630 ms631 ms663 ms789 ms
20 September 2026619 ms725 ms—845 ms
21 September 2026700 ms638 ms1138 ms722 ms
22 September 2026638 ms637 ms814 ms881 ms

Does your location matter?

DeepSeek direct was faster from Singapore and Mumbai in our tests.

Typical time to first token by city: the median of every daily request from that city over the window.
050010001500 msDeepSeek direct from Amsterdam: 733 ms typical time to first token; slower requests 698 ms.OpenRouter · Cloudflare from Amsterdam: 537 ms typical time to first token; slower requests 1027 ms.OpenRouter · Together from Amsterdam: 722 ms typical time to first token; slower requests 1064 ms.OpenRouter · DigitalOcean from Amsterdam: 608 ms typical time to first token; slower requests 925 ms.AmsterdamDeepSeek direct from San Francisco: 781 ms typical time to first token; slower requests 712 ms.OpenRouter · Cloudflare from San Francisco: 659 ms typical time to first token; slower requests 1830 ms.OpenRouter · Together from San Francisco: 627 ms typical time to first token; slower requests 2772 ms.OpenRouter · DigitalOcean from San Francisco: 546 ms typical time to first token; slower requests 1435 ms.San FranciscoDeepSeek direct from Montreal: 732 ms typical time to first token; slower requests 653 ms.OpenRouter · Cloudflare from Montreal: 643 ms typical time to first token; slower requests 880 ms.OpenRouter · Together from Montreal: 692 ms typical time to first token; slower requests 4746 ms.OpenRouter · DigitalOcean from Montreal: 422 ms typical time to first token; slower requests 571 ms.MontrealDeepSeek direct from Singapore: 621 ms typical time to first token; slower requests 590 ms.OpenRouter · Cloudflare from Singapore: 782 ms typical time to first token; slower requests 1295 ms.OpenRouter · Together from Singapore: 967 ms typical time to first token; slower requests 1145 ms.OpenRouter · DigitalOcean from Singapore: 939 ms typical time to first token; slower requests 1763 ms.SingaporeDeepSeek direct from Tokyo: 575 ms typical time to first token; slower requests 621 ms.OpenRouter · Cloudflare from Tokyo: 784 ms typical time to first token; slower requests 1312 ms.OpenRouter · Together from Tokyo: 724 ms typical time to first token; slower requests 1206 ms.OpenRouter · DigitalOcean from Tokyo: 1081 ms typical time to first token; slower requests 1598 ms.TokyoDeepSeek direct from Mumbai: 664 ms typical time to first token; slower requests 516 ms.OpenRouter · Cloudflare from Mumbai: 1211 ms typical time to first token; slower requests 1390 ms.OpenRouter · Together from Mumbai: 1183 ms typical time to first token; slower requests 5256 ms.OpenRouter · DigitalOcean from Mumbai: 1348 ms typical time to first token; slower requests 3380 ms.Mumbai
  • DeepSeek direct
  • OpenRouter · Cloudflare
  • OpenRouter · Together
  • OpenRouter · DigitalOcean
Show exact values
CityDeepSeek directOpenRouter · CloudflareOpenRouter · TogetherOpenRouter · DigitalOcean
Amsterdam733 msslower 698 ms537 msslower 1.0 s722 msslower 1.1 s608 msslower 925 ms
San Francisco781 msslower 712 ms659 msslower 1.8 s627 msslower 2.8 s546 msslower 1.4 s
Montreal732 msslower 653 ms643 msslower 880 ms692 msslower 4.7 s422 msslower 571 ms
Singapore621 msslower 590 ms782 msslower 1.3 s967 msslower 1.1 s939 msslower 1.8 s
Tokyo575 msslower 621 ms784 msslower 1.3 s724 msslower 1.2 s1081 msslower 1.6 s
Mumbai664 msslower 516 ms1211 msslower 1.4 s1183 msslower 5.3 s1348 msslower 3.4 s

Bold is the lowest typical time in that city. "Slower" is the time 95% of requests in the controlled run finished within.

Where the waiting happens

How much of the wait a closer server could remove, and how much is the provider’s own queue and model.

Median city in each provider’s controlled 20-request run (7 September 2026; 13 September 2026; 20 September 2026). Both bars share one scale.
  • DeepSeek directthe network is about 2% of the wait
    Network setup
    11 ms
    Waiting for the model
    451 ms

    First visible token after 462 ms in total.

  • OpenRouter · Cloudflarethe network is about 2% of the wait
    Network setup
    18 ms
    Waiting for the model
    717 ms

    First visible token after 735 ms in total.

  • OpenRouter · Togetherthe network is about 2% of the wait
    Network setup
    19 ms
    Waiting for the model
    779 ms

    First visible token after 798 ms in total.

  • OpenRouter · DigitalOceanthe network is about 2% of the wait
    Network setup
    18 ms
    Waiting for the model
    865 ms

    First visible token after 883 ms in total.

Network setup is finding the server, reaching it and setting up a secure connection. Waiting for the model is everything after that until the first visible token: queueing and the model’s own work.

Side by side

9 September 2026 to 22 September 2026DeepSeek directOpenRouter · CloudflareOpenRouter · TogetherOpenRouter · DigitalOcean
Modeldeepseek-v4-flashdeepseek-v4-flash-0731deepseek-v4-flash-0731deepseek-v4-flash-0731
Typical time to first token (median city)698 ms721 ms723 ms774 ms
Slower requests, p95 (median city)637 ms1304 ms1989 ms1517 ms
Fastest cityTokyo, 575 msAmsterdam, 537 msSan Francisco, 627 msMontreal, 422 ms
Slowest citySan Francisco, 781 msMumbai, 1211 msMumbai, 1183 msMumbai, 1348 ms
Spread between cities36%126%89%219%
Days when every city slowed together0010
Requests in this window420 over 14 daily runs417 over 14 daily runs326 over 11 daily runs419 over 14 daily runs
Latest controlled run7 September 202620 September 202613 September 202620 September 2026

How we test

We send the same streaming prompt from six cities to DeepSeek directly and through OpenRouter, then measure time to first visible token.

Same request on both sides: thinking off, temperature 0. DeepSeek direct has one controlled run so far. The Fireworks and BaseTen routes are left out: one throttled us, the other never returned a visible token.

Full testing methodology →

How does your API compare?

Test your endpoint from the same 6 cities used here and see your response time next to these numbers.

Run a free latency test

No account required · Takes about 30 seconds.

Explore each API