Skip to content
Measuring continuouslyEthereumBSCSolana

Which RPC provider is actually fastest from where you are?

RPCNode.IO continuously measures Ethereum, BNB Smart Chain and Solana RPC providers from distributed probes — latency, availability, response correctness and chain-head freshness, per method and per region.

Providers
0
Methods
0
Chains
0
Regions
0

Collecting right now

Connecting

0

measurements stored in the last 24 hours

  • Waiting for the next measurement…
Open the measurement lab

Live

Collecting right now

Every row below is a call that was made, timed and stored. The feed is pushed as each measurement lands rather than polled, so what you are watching is the collection pipeline itself.

How it is measured
Connecting

Measurements arriving from 0 probe regions

0 providers · 0 methods · 0 chains

Observations / 24h

0

last known figure

Fastest just now

waiting for a measurement

Median, live window

waiting for a measurement

Leading on latency

needs five samples from one provider

Measurement tape

newest first

  • Waiting for the next measurement…

Figures on this panel cover the last few minutes of collection only. Every ranking elsewhere on the site is computed over a stated window from merged percentile states — the two are not the same measurement. Methodology

Live

Every chain, right now

Each card carries that chain's current top three by p95 response time, over the last 24 hours.

All chains

Ethereum

~12s blocks · chain id 1

Providers
9
Endpoints
12
Validators
400K

No ranking yet for this chain. A provider needs a minimum sample count before it is scored at all.

BSC

~450ms blocks · chain id 56

Providers
5
Endpoints
8
Validators
0

No ranking yet for this chain. A provider needs a minimum sample count before it is scored at all.

Solana

~420ms blocks

Providers
3
Endpoints
5
Validators
0

No ranking yet for this chain. A provider needs a minimum sample count before it is scored at all.

Methods

Where providers actually differ

They are broadly equivalent on cheap lookups and diverge sharply on the expensive ones. These are the calls where the difference shows, and what each one is for.

Full support matrix

Methodology

How a number gets here

Every published figure is traceable to collected observations. Where the data is thin, the interface says so rather than guessing.

Read the methodology
  1. 01

    Probes call the real endpoint

    Measurement runs from probe agents rather than from this server, over the same public network path a real client would take. Each probe holds its own credential and its own rate budget.

  2. 02

    Every timing is recorded separately

    DNS, TCP, TLS, time to first byte and download are stored as distinct fields. A provider that resolves slowly and one that computes slowly look identical in a single total, and they are not the same problem.

  3. 03

    Failures are classified, not counted

    An unsupported method is a capability fact. A rate limit is a provider protecting itself. A bug in our own software is never counted against anyone. Collapsing these into one error rate would be the easiest way to publish a wrong ranking.

  4. 04

    Correctness is checked against the cohort

    Responses are validated for envelope, type and chain, and the chain head is compared against a robust median across every provider measured at that instant — never against one provider treated as truth.

Guides

Or run it yourself

The rankings answer which endpoint to call. These answer the questions that come before that, and the ones that come after deciding not to call anybody else's — general where the problem is general, and per chain where it is not.

All 11 articles

What this does not do

  • No paid placement. Rankings follow the measurements. A provider cannot buy a position, and none has been asked to.
  • No invented data. Benchmark results, validator identities, endpoint capabilities and geographic locations are published only where they were observed. Unknown stays unknown.
  • No write calls. Nothing here signs or sends a transaction. Every measured method is a read.
  • No hammering. Endpoints are called at conservative rates with per-endpoint budgets and backoff, because they are other people’s infrastructure.

Take the data

Every ranking, method result and validator record on this site is served by a versioned public REST API. Same numbers, same windows, no key required.

curl https://api.rpcnode.io/api/v1/rankings?chain=ethereum
API reference