Crypto Exchange Feed Latency: 16 Venues Measured

Vultax maintains order-book snapshot connections to a set of centralised crypto exchanges. Each snapshot carries an observed latency measurement taken at capture time. This article reports those measurements directly, without modelling or interpolation.

The latency figures cover a one-hour sample of order-book snapshots. The uptime figures cover the 24-hour period ending 2026-09-02 00:00 UTC. Both windows are stated because they are short, and a reader should weigh the result against the sample behind it.

• Latency — per-snapshot age recorded at capture, reported as the median (p50) and the 95th percentile (p95)
• Uptime — distinct one-minute buckets containing at least one snapshot, out of 1,440
• No gap filling — missing intervals are reported as missing, never carried forward or interpolated

Medians cluster tightly. Fourteen of the sixteen venues sit between 69 ms and 93 ms, with Coinbase the outlier on the fast side and BTSE on the slow side. The full ordering at p50:

• Coinbase 35.0 ms · Bitrue 69.5 ms · Bybit 72.8 ms · WhiteBIT 72.8 ms · Gemini 75.4 ms
• Binance 76.8 ms · Gate.io 77.2 ms · BitMEX 78.0 ms · Poloniex 79.8 ms · OKX 80.0 ms
• HTX 80.4 ms · Bitget 81.0 ms · KuCoin 81.0 ms · Bitstamp 82.9 ms · Upbit 92.8 ms · BTSE 111.5 ms

The 95th percentile behaves very differently from the median. It ranged from 288.5 ms to 589.7 ms across the same sixteen venues, and the ordering at p95 does not match the ordering at p50: Gemini, fifth-fastest by median, drops to thirteenth, while BTSE, slowest by median, is not the slowest tail. A latency figure quoted as a single number is almost always a median, and the median is the flattering half of the distribution. The full ordering at p95:

• Coinbase 288.5 ms · Bitrue 311.6 ms · HTX 324.3 ms · OKX 326.5 ms · Binance 328.1 ms · Bitget 330.7 ms
• WhiteBIT 330.9 ms · Bybit 331.1 ms · KuCoin 332.5 ms · Gate.io 333.5 ms · Poloniex 338.6 ms · BitMEX 338.9 ms
• Gemini 361.8 ms · BTSE 383.5 ms · Bitstamp 406.0 ms · Upbit 589.7 ms

Across the 24-hour window, all sixteen venues returned data in every one of the 1,440 one-minute buckets. On the evidence of this window, continuity of connection was not what separated them; latency was.

Kraken is not in this ranking. Its connector moved to full local order-book reconstruction on 2026-09-02, and the sample collected before that change (154 snapshots) is too small to rank against venues with hundreds of thousands. The mechanism is described in why a connected exchange feed can record nothing; Kraken is included in the next measurement window.

Vultax describes its latency as a sub-100ms target. This measurement is consistent with that wording and does not support strengthening it. Fifteen of sixteen venues sit under 100 ms at the median, but every venue exceeds 100 ms at p95, and the slowest tails run past half a second. For anyone comparing prices across two venues, the tail is the operative number, which is why most displayed arbitrage spreads are not executable.

Feed latency is also only one component of end-to-end latency. It measures the age of a snapshot at capture against the venue, not the time for a signal to reach a screen. Readers comparing vendors should confirm which of those any published figure refers to, because the two differ by a wide margin and are frequently quoted interchangeably.

The strip at the top of this page is not a screenshot of the day this was written. Every ten minutes Vultax re-reads five venues' status pages, the trade tapes of five books, the five order books, and CoinGecko's exchange tickers and rewrites the figures; the Vi IQ beneath them scores the same six domains the terminal scores for a pair, computed for BTC on the major spot venues, with any domain that cannot be computed shown as unavailable rather than filled in. Read the numbers as a live check on the argument above, and the revision notes at the end for what has changed since publication.

Latency is measured here on Vultax's own collectors, so the strip shows the conditions the collectors are measuring rather than a second latency figure: how many venues report healthy, the cross-venue gap between mids, the mean best-quote spread, and the number of large prints in the last half hour. A venue that is 'healthy' on its status page and silent on its tape is the case the coverage measurement in this article exists to catch.

Two public outages since this was written show what the tail of the distribution looks like in practice. In the 20 October 2025 AWS outage Coinbase was unreachable while Binance, Kraken and OKX stayed up; Coinbase's postmortem of 7 May 2026 describes roughly eight hours down and about twelve more to full recovery after an AWS thermal event. Neither shows in a median; both show in the 95th percentile and in coverage.

Latency claims elsewhere have become a marketing line. Prediction-market whale trackers advertise alerts 'in under two seconds' and typical figures of five to fifteen seconds, and none states its measurement method. The point of publishing a p50 and a p95 over a stated window across sixteen venues is that the number can be checked, and the limits section below says exactly what it does not cover.

This is a short-window observation from a single collection point, not a long-run reliability study. Latency measured from one vantage reflects the network path between that point and each venue as much as it reflects the venue, so the ordering could change from other infrastructure. Independent monitors that probe exchanges from many regions, such as Glassnode's latency monitor, report WebSocket round-trip time — a different quantity from the snapshot age reported here, and not directly comparable to it.

• Windows — one hour for latency, 24 hours for uptime
• Retention — raw order-book data is kept for five days, which bounds how far back these figures can be recomputed
• Scope — nothing here is a statement about any exchange's internal matching performance, and no claim is made about wrongdoing or the conduct of any venue or account holder