Everyone claims to be fast. This is a node in Tokyo asking each endpoint for the head of the chain, six times a run, and writing down what came back — public endpoints, protected endpoints, and ours in the same table without favour.
| Endpoint | Kind | Now | 24 h median | Blocks behind | Failed calls |
|---|---|---|---|---|---|
| Loading the measurement… | |||||
Response time is how long eth_blockNumber took from Tokyo. Measured from Frankfurt or São Paulo the order would change, and a table that does not say where it was measured is telling you about its own location. Read it as a floor, not as what you will see.
Blocks behind is how far behind the current head that endpoint answered, and it travels: an endpoint two blocks behind gives you a price that is almost a second old wherever you are sitting, which is how a swap quotes at one number and fills at another. On a chain with 0.45-second blocks, this is the number that costs money.
Failed calls counts what did not come back at all over the day — timeouts, rate limits, errors. A fast endpoint that refuses one call in twenty is slower than a steady one.
Reading the chain is the easy half. Every endpoint here answers questions well; what separates them is what happens to a transaction you send — whether it goes into the public mempool where anyone can trade against it, or straight to the builders who assemble the block. That cannot be measured with a stopwatch from outside, so it is not in this table. It is the whole of what we do, and it is described on the front page rather than smuggled into a benchmark.
We run one of the endpoints listed. We publish the table anyway, including the runs where we are not first — a measurement you can only trust when it flatters its author is not a measurement. The raw JSON behind it is at rpc.backrunshield.com/bench; take it and check.