DEEPSEEK LATENCY BENCHMARK · Oct 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 527–589 ms, routes 879–1,339 ms
    • Montreal: routes 416–926 ms, direct 734 ms
    See the chart by city →

  2. Finding 2

    DeepSeek's direct API was more consistent

    DeepSeek's API varied less between locations: its slowest city was 39% above its fastest, against 81–208% through OpenRouter.

    • City spread: 39% direct, 81–208% via OpenRouter
    • Slow days: 0 direct, 0 via OpenRouter
    See the day-by-day chart →

  3. Finding 3

    Most of the wait isn't the network

    Connection setup was only 14–15 ms. Most latency came after reaching the provider.

    • Network: 14–15 ms
    • First token: 693–1,030 ms
    See where the time goes →

Time to first token, day by day

Daily median time to first token, 11 September 2026 to 10 October 2026. Each point is the median of the compared cities’ daily medians, 5 requests per city.
0200040006000 ms11 Sept10 Oct
Show the numbers
DayDeepSeek directOpenRouter · TogetherOpenRouter · DigitalOceanOpenRouter · Cloudflare
11 September 2026611 ms697 ms705 ms729 ms
12 September 2026645 ms859 ms736 ms873 ms
13 September 2026676 ms689 ms843 ms650 ms
14 September 2026802 ms757 ms792 ms748 ms
15 September 2026902 ms687 ms752 ms1274 ms
16 September 2026733 ms—695 ms714 ms
17 September 2026594 ms—861 ms745 ms
18 September 2026732 ms5566 ms806 ms727 ms
19 September 2026630 ms663 ms789 ms631 ms
20 September 2026619 ms—845 ms725 ms
21 September 2026700 ms1138 ms722 ms638 ms
22 September 2026638 ms814 ms881 ms637 ms
23 September 2026667 ms1772 ms946 ms—
24 September 2026647 ms1074 ms729 ms774 ms
25 September 2026558 ms2822 ms826 ms602 ms
26 September 2026655 ms1120 ms871 ms658 ms
27 September 2026699 ms776 ms651 ms784 ms
28 September 2026641 ms792 ms752 ms1046 ms
29 September 2026662 ms751 ms845 ms904 ms
30 September 2026700 ms713 ms866 ms—
1 October 2026647 ms—844 ms916 ms
2 October 2026685 ms716 ms833 ms906 ms
3 October 2026711 ms618 ms772 ms699 ms
4 October 2026602 ms531 ms865 ms792 ms
5 October 2026648 ms1202 ms846 ms—
6 October 2026657 ms4123 ms813 ms939 ms
7 October 2026654 ms944 ms770 ms986 ms
9 October 2026672 ms820 ms805 ms—
10 October 2026558 ms585 ms817 ms1509 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: 700 ms typical time to first token; slower requests 918 ms.OpenRouter · Together from Amsterdam: 779 ms typical time to first token; slower requests 4623 ms.OpenRouter · DigitalOcean from Amsterdam: 583 ms typical time to first token; slower requests 970 ms.OpenRouter · Cloudflare from Amsterdam: 779 ms typical time to first token; slower requests 1894 ms.AmsterdamDeepSeek direct from San Francisco: 728 ms typical time to first token; slower requests 936 ms.OpenRouter · Together from San Francisco: 582 ms typical time to first token; slower requests 5346 ms.OpenRouter · DigitalOcean from San Francisco: 574 ms typical time to first token; slower requests 955 ms.OpenRouter · Cloudflare from San Francisco: 691 ms typical time to first token; slower requests 2036 ms.San FranciscoDeepSeek direct from Montreal: 734 ms typical time to first token; slower requests 990 ms.OpenRouter · Together from Montreal: 626 ms typical time to first token; slower requests 6193 ms.OpenRouter · DigitalOcean from Montreal: 416 ms typical time to first token; slower requests 741 ms.OpenRouter · Cloudflare from Montreal: 926 ms typical time to first token; slower requests 1679 ms.MontrealDeepSeek direct from Singapore: 527 ms typical time to first token; slower requests 752 ms.OpenRouter · Together from Singapore: 879 ms typical time to first token; slower requests 4825 ms.OpenRouter · DigitalOcean from Singapore: 1008 ms typical time to first token; slower requests 1973 ms.OpenRouter · Cloudflare from Singapore: 1016 ms typical time to first token; slower requests 2133 ms.SingaporeDeepSeek direct from Tokyo: 622 ms typical time to first token; slower requests 846 ms.OpenRouter · Together from Tokyo: 678 ms typical time to first token; slower requests 3923 ms.OpenRouter · DigitalOcean from Tokyo: 1084 ms typical time to first token; slower requests 2072 ms.OpenRouter · Cloudflare from Tokyo: 968 ms typical time to first token; slower requests 2298 ms.TokyoDeepSeek direct from Mumbai: 589 ms typical time to first token; slower requests 834 ms.OpenRouter · Together from Mumbai: 1055 ms typical time to first token; slower requests 3566 ms.OpenRouter · DigitalOcean from Mumbai: 1280 ms typical time to first token; slower requests 1970 ms.OpenRouter · Cloudflare from Mumbai: 1339 ms typical time to first token; slower requests 2681 ms.Mumbai
  • DeepSeek direct
  • OpenRouter · Together
  • OpenRouter · DigitalOcean
  • OpenRouter · Cloudflare
Show exact values
CityDeepSeek directOpenRouter · TogetherOpenRouter · DigitalOceanOpenRouter · Cloudflare
Amsterdam700 msslower 918 ms779 msslower 4.6 s583 msslower 970 ms779 msslower 1.9 s
San Francisco728 msslower 936 ms582 msslower 5.3 s574 msslower 955 ms691 msslower 2.0 s
Montreal734 msslower 990 ms626 msslower 6.2 s416 msslower 741 ms926 msslower 1.7 s
Singapore527 msslower 752 ms879 msslower 4.8 s1008 msslower 2.0 s1016 msslower 2.1 s
Tokyo622 msslower 846 ms678 msslower 3.9 s1084 msslower 2.1 s968 msslower 2.3 s
Mumbai589 msslower 834 ms1055 msslower 3.6 s1280 msslower 2.0 s1339 msslower 2.7 s

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

Where does the wait happen?

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

Both bars share one scale. They come from one test of 20 requests per city on 27 September 2026 and 4 October 2026, not from the two weeks of daily requests behind the typical times, so they won't match those exactly.
  • DeepSeek direct
    Network setup
    14 ms
    Waiting for the model
    679 ms

    First visible token after 693 ms in total.

  • OpenRouter · Together
    Network setup
    14 ms
    Waiting for the model
    722 ms

    First visible token after 736 ms in total.

  • OpenRouter · DigitalOcean
    Network setup
    15 ms
    Waiting for the model
    848 ms

    First visible token after 863 ms in total.

  • OpenRouter · Cloudflare
    Network setup
    14 ms
    Waiting for the model
    1016 ms

    First visible token after 1030 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.

Direct vs OpenRouter

27 September – 10 October 2026 · six test locations

MetricDirectTogetherDigitalOceanCloudflare
Typical TTFT661 ms729 ms796 ms947 ms
Location spread39%81%208%94%

Direct uses deepseek-v4-flash; the OpenRouter routes use deepseek-v4-flash-0731. Methodology →

Show full benchmark details ↓
DetailDeepSeek directOpenRouter · TogetherOpenRouter · DigitalOceanOpenRouter · Cloudflare
Modeldeepseek-v4-flashdeepseek-v4-flash-0731deepseek-v4-flash-0731deepseek-v4-flash-0731
Slower requests, p95 (median city)882 ms4724 ms1470 ms2085 ms
Fastest locationSingapore · 527 msSan Francisco · 582 msMontreal · 416 msSan Francisco · 691 ms
Slowest locationMontreal · 734 msMumbai · 1055 msMumbai · 1280 msMumbai · 1339 ms
Days when every city slowed together0000
Requests in this window390 over 13 daily runs355 over 12 daily runs390 over 13 daily runs300 over 10 daily runs
Latest controlled run27 September 20264 October 20264 October 20264 October 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. The Fireworks and BaseTen routes are left out: one throttled us, the other never returned a visible token.

Full testing methodology →

Good to know

  • Network setup is the hop to OpenRouter's edge (14–15 ms from every city); the hop from OpenRouter to the provider's machines is counted as waiting for the model, which is why Mumbai still waits longer than Montreal.
  • Direct sends deepseek-v4-flash, which DeepSeek has served with V4.1 Flash since 10 September (V4 Flash is retired), while the OpenRouter routes pin deepseek-v4-flash-0731, so the two sides have not been the same model since that day.
  • Each OpenRouter request pins one provider with fallbacks off, and a city's run is discarded if any response came from another provider.
  • The prompt is "Say 'ok' and nothing else." (max_tokens 256, temperature 0, thinking off), sent daily between 03:41 and 07:18 UTC, 5 requests per city after one discarded warm-up.
  • We time the first visible token only; tokens per second, long prompts and thinking on are not measured.

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