Trading API Latency: What a Ping Really Tells You (Our Data)

We pinged the trading APIs of Polymarket, Kalshi, Hyperliquid, Binance and Coinbase from all six of our data centers to see what trading API latency looks like from each city. From Amsterdam, Frankfurt, London and New York, nearly every hostname replied in roughly 2 milliseconds or less, including venues whose servers run on another continent. That's the headline: a ping to a trading API mostly tells you how far away the nearest content delivery network (CDN) edge is, not the exchange. Below are the numbers, what they can and can't tell you, and a more reliable way to measure before you pick a home for your bot.
How we measured
Tool: the free latency checker on our site. It fires ICMP pings from probes that live in each of the six OrbitServers facilities: Amsterdam, London, Frankfurt, New York, Salt Lake City and Tokyo.
Targets: hostnames the venues publish in their own API docs for trading and market data.
Timing: three passes on 8 November 2026, starting around 18:20, 18:30 and 18:41 UTC.
Vantage point: the probes are looking-glass nodes inside our data centers, not customer VPS instances.
Numbers shown: the middle value of the three passes, in milliseconds. Individual passes can jump around: Frankfurt to Hyperliquid read 106.79 ms once, yet both other passes came in below 1 ms.
A ping covers the network path only. How quickly a venue fills or confirms your order is a separate question it can't answer.
Trading API latency results
Median round-trip time in milliseconds, measured 8 November 2026.
| API hostname | Amsterdam | London | Frankfurt | New York | Salt Lake City | Tokyo |
|---|---|---|---|---|---|---|
clob.polymarket.com | 1.05 | 1.86 | 1.23 | 1.75 | 1.27 | 0.45 |
ws-subscriptions-clob.polymarket.com | 0.44 | 0.96 | 0.99 | 1.76 | 10.79 | 89.08 |
gamma-api.polymarket.com | 0.47 | 1.07 | 0.89 | 3.01 | 10.8 | 92.37 |
api.elections.kalshi.com | 0.57 | 0.89 | 0.9 | 1.75 | 15.83 | 92.78 |
api.kalshi.com | no reply | no reply | no reply | no reply | no reply | no reply |
external-api.kalshi.com | no reply | no reply | no reply | no reply | no reply | no reply |
api.hyperliquid.xyz | 0.57 | 1.02 | 0.78 | 2.2 | 10.89 | 1.34 |
api.binance.com | 0.37 | 0.76 | 0.72 | 2.14 | 11 | 1.55 |
api.coinbase.com | 1.16 | 1.68 | 1.18 | 1.57 | 1.17 | 0.68 |
What the numbers show
Most API hostnames answer from a nearby edge
A sub-millisecond ping from Amsterdam to a service hosted far away isn't possible over that distance. Light in fiber takes much longer than that just to cross an ocean. The reply comes from a CDN or load-balancing edge in or near the same city, and that edge passes your request on to the exchange's own servers.
It also explains why clob.polymarket.com and api.coinbase.com came back in under 2 ms from every location, Salt Lake City and Tokyo included: both seem to have edges near each of our data centers.
Some edges are missing in some regions
In some regions, though, the first hop that answers is far off. We measured between 89.08 and 92.78 ms from Tokyo to Polymarket's WebSocket and Gamma hosts and Kalshi's api.elections.kalshi.com; Amsterdam reached the same three in under 1 ms. Salt Lake City sat between 10.79 and 15.83 ms on several hosts, where New York never went above 3.01 ms.
The pattern flips for Hyperliquid and Binance: 1.34 and 1.55 ms from Tokyo, yet 10.89 and 11 ms from Salt Lake City.
Some hosts don't answer pings at all
Two Kalshi hosts, api.kalshi.com and external-api.kalshi.com, stayed silent everywhere. That's normal for cloud load balancers, which commonly throw ICMP away; silence says nothing about whether the API is up. A lookup on 7 November 2026 placed Kalshi's main API host in AWS us-east-2 (Ohio), something our Kalshi API tutorial covers too.
Why this matters for a trading bot
An order still has to go from the edge to the exchange's matching engine and back again. Seeing a 0.57 ms ping from Amsterdam to a Kalshi hostname doesn't mean an Amsterdam server is 0.57 ms from Kalshi's servers in Ohio. Pick a location on ping alone and you can end up on the wrong continent.
How to measure real API latency
Instead of pinging, time a real HTTPS request. curl can split one request into its separate stages:
curl -o /dev/null -s -w \
"dns %{time_namelookup}s tcp %{time_connect}s tls %{time_appconnect}s first byte %{time_starttransfer}s\n" \
"https://clob.polymarket.com/time"
tcp and tls: the handshake times, which reveal how close the box ending your connection is. Most of the time that's the edge.
first byte: adds the hop to whatever server really produces the answer, and that server's thinking time. When the edge can't reply from its cache, this figure is the closest stand-in for what a bot will see.
One run proves little. Loop it 20 to 50 times on each server you're weighing, then look at the median and the worst case:
for i in $(seq 1 30); do
curl -o /dev/null -s -w "%{time_starttransfer}\n" "https://clob.polymarket.com/time"
done | sort -n | awk '{a[NR]=$1} END {print "median", a[int(NR/2)+1], "max", a[NR]}'
Replace the URL with an endpoint your bot actually calls, an order book request for example, and compare servers in different cities. If you stream live data, also time how long WebSocket messages take to arrive, since that's what a market-making loop responds to.
Picking a location for each venue
Before latency, look at what each venue permits:
Polymarket. Each API call is judged by its source IP. The docs say trading is close-only from the United States, the United Kingdom, Germany and other countries, while the Netherlands and Japan are restricted on the website only, not the API. Hit
GET https://polymarket.com/api/geoblockfrom your box first, and remember that the rules bind you personally wherever the server is. Endpoint by endpoint details live in our Polymarket API guide.Kalshi. Access follows your verified account rather than the machine. With the main API host sitting in AWS us-east-2 at the time of our check, a New York VPS is the nearest OrbitServers location. Those hosts won't answer pings, so Kalshi latency has to be judged with HTTP timing.
Anything else. Use the curl timing from two or three cities you're considering, then decide.
Run the test yourself
Anyone can use the latency checker, and it takes any public hostname, so you can repeat this test for whichever API you trade on from all six cities. Then host the bot wherever the data points. OrbitServers VPS plans start at $39.99/mo, deploy automatically once payment is confirmed, and are offered in every city above. Solana traders can find block engine numbers on our Jito latency benchmarks page.
Frequently asked questions
Why is my ping to an exchange API so low?
Nearly all trading APIs sit behind a CDN or a global load balancer. The nearest edge answers your ping, and it may be in your own city, far from the exchange's servers.
Why does Polymarket latency look the same from every city?
Because clob.polymarket.com replies from a nearby edge almost everywhere, the ping only shows how far that edge is. To judge Polymarket latency for trading, check the API's location rules first, then time real requests.
Why doesn't Kalshi's API answer pings?
Neither api.kalshi.com nor external-api.kalshi.com answered ICMP from any of our locations. Cloud load balancers often drop pings, so use an HTTPS request to measure instead.
What is the best way to measure trading API latency?
Time real HTTPS requests with curl, run them many times from every server you're considering, and compare medians and worst cases. If you consume streaming data, measure WebSocket message delay too.
Does lower latency guarantee better fills?
No. Latency is only one piece. Strategy, fees, rate limits and the venue's own processing count just as much, and network conditions shift over time.
Get started with Orbit Servers
Low-latency VPS, bare metal, and colocation across the US, EU, and APAC - provisioned instantly and built for performance-critical workloads.
Get startedRelated products
Written by
Julius
Product Manager
Julius is a product-minded builder who cares about making technology simpler and more useful for the people who rely on it. He’s always looking for ways to improve things, one small step at a time, and enjoys turning good ideas into products that work.