built 2026-10-06 05:12 UTC · commit b84c6a4cbafc

RequestBench Results

RequestBench measures 38 HTTP frameworks in .NET, Go, Java, Node, Python and Rust. They run on 4 hosts: in a container over HTTP/1.1, in a container over HTTP/2, as a Lambda function with 512 MB and as a Lambda function with 1,024 MB. Every framework answers the same 65 tests, which cover the features a web service relies on, from JSON and validation to CORS, caching and compression.

Loading the results catalog…

Over time

The top six rows of the current ranking, over the runs on one machine with the newest run's rate ladder and corpus. The same framework on another CPU can take twice as long, so two machines never share a chart.

Each row is the time that framework took at the offered rate, on the machine and in the window named above. A framework that could not sustain a rate has no latency at it: percentiles over the requests that survived would flatter the collapse, so what it achieved and what it dropped is shown instead. A hatched number is thin: fewer than 10 of the requests it is read from lie beyond it.