RPC performance,
measured in the open

Quicknode runs synthetic probes from 15 hosts across three clouds, AWS, Google Cloud, and Oracle Cloud, in five regions: N. Virginia, California, Frankfurt, Tokyo, and Singapore. Every provider receives identical payloads on the same schedule, and each probe opens a fresh connection. Latency reflects successful responses only, and the full methodology is published.

P95 latency is the value that 95 percent of requests come in under, so it captures the slow tail that averages hide. It matters because real users feel the worst requests, not the median, and a provider with a low average but a high P95 can still cause timeouts under load. When choosing an RPC provider, P95 is usually the most telling single number.

P50 is the median: half of all requests finish faster than this number and half finish slower. It describes a typical request, but on its own it hides the slow tail, which is why it is best read next to P95.

The average is the mean of every measured request in the window. A few slow responses pull it above the median, so it can read higher than a typical request. Comparing the average with P50 shows how skewed the results are.

The benchmarks compare Quicknode against providers including Alchemy, Infura, Ankr, Chainstack, Google, and NodeReal on EVM chains, plus Solana specialists Helius and Triton. The exact set shown depends on which providers are measured for the selected chain and region.

Latency is measured on Ethereum, Base, Arbitrum, Optimism, BNB Smart Chain, Polygon, and Solana, with an EVM aggregate view. You can filter by any of five regions: N. Virginia, California, Frankfurt, Tokyo, and Singapore, or by Global, which averages the five regions with equal weight so no single region dominates the number. More chains, including Hyperliquid, are being added.

The dashboard fetches live numbers from the QuickLee API when the page loads and when you return to the tab. Reload anytime for a hard refresh. QuickLee reports in time buckets that widen with the window, from under a minute on the 1-hour view to roughly 15 minutes on the 24-hour view, so two quick refreshes can show the same digits until the next bucket lands. The freshness chip shows how long it has been since the last successful fetch, and keeps climbing until the next one.

Warm, reused connections skip the TCP and TLS handshake and make a provider look faster than a real first request. Quicknode opens a fresh connection for every probe, so the numbers reflect what a real application pays rather than a best case keep-alive.

Failures, timeouts, and rate limits are recorded as-is. A failed request is never replaced by retrying until it succeeds, so failures stay visible in the results. The latency figures themselves reflect successful responses only, which keeps the comparison honest instead of rewarding providers that silently drop requests.

Compare providers on the chain and region closest to your users, and weigh P95 alongside the average so the slow tail is accounted for. Then confirm the provider supports the methods and archive data your workload needs. You can start on the Quicknode free tier and test against these numbers directly.