Performance

Illustrative RPC throughput and latency from a controlled multi-account EVMKit benchmark.

This page captures representative EVMKit benchmark results from controlled load-test runs. Treat them as directional references, not guarantees.

Actual results vary with request mix, CU weight, region, account configuration, connection reuse, and current limiter state.

Different RPC methods consume CU differently and exercise different execution paths, so results measured with evmkit_call may differ from what you see with other methods.

RPC benchmark

Load shape

reference date: 2026-03-22
duration: 180s
accounts: 10
transport: RPC over HTTP/2
RPC target rate: 10,000 req/s per account
RPC connections: 64 persistent HTTP/2 connections per account (16 per endpoint)
regions under test: us1, us2, eu1, eu2
regional load hosts: US regions were tested from a nearby US load server and EU regions from a nearby EU load server, but not from the same machines that serve the public endpoints
RPC payload: evmkit_call for WETH.balanceOf(Vitalik) with createOutput=true and createAccessList=true
assumed successful RPC cost: 1 CU

RPC results

Metric	Value
`200` responses	`17,646,141`
`434` responses	`353,767`
Success ratio	`98.03%`
`434` ratio	`1.97%`
Aggregate offered throughput	`99,999.49 req/s`
Aggregate successful throughput	`98,034.12 req/s`
Aggregate successful CU rate	`98,034.12 CU/s`
Successful throughput per account	about `9.8k req/s`

RPC latency percentiles

Response set	p50	p95	p99
All responses	`0.580 ms`	`1.410 ms`	`2.330 ms`
Successful responses	`0.583 ms`	`1.420 ms`	`2.350 ms`

RPC regional breakdown

Endpoint	Successful throughput	p50	p95	p99
`us1`	`24,580.59 req/s`	`0.500 ms`	`1.320 ms`	`2.580 ms`
`us2`	`24,590.88 req/s`	`0.502 ms`	`1.380 ms`	`2.680 ms`
`eu1`	`24,431.66 req/s`	`0.655 ms`	`1.440 ms`	`2.190 ms`
`eu2`	`24,430.99 req/s`	`0.659 ms`	`1.470 ms`	`2.220 ms`

Interpretation

The aggregate figures are 10-account results, not single-account results.
This run pushed slightly beyond the fully successful short-window operating range, so some 434 responses were expected.
In this benchmark, 434 indicates normal short-window rate-limit enforcement rather than service instability. See Rate limits.

Rate limits for 434, Retry-After, and transport guidance
Regions for regional hostnames and HTTP/2 notes