Laravel Octane Benchmark: FrankenPHP vs Swoole vs RoadRunner vs PHP-FPM

Most Octane benchmarks are a hello-world on hardware nobody describes. I ran the same Laravel app under PHP-FPM, FrankenPHP, Swoole and RoadRunner in Docker on one laptop, three runs per cell, then broke it on purpose in the ways Octane breaks apps. You get the numbers (laptop numbers, with their spread), every Dockerfile needed to rerun them, the failures with real output, and the fixes.

Laravel Octane benchmark setup

• App: fresh laravel/laravel with laravel/octane v2.21.0, which the Octane releases page showed as the latest release (06 Oct) when I checked on 11 Oct 2026. Default web middleware, array session and cache drivers, SQLite.
• Versions as reported inside my containers (I did not cross-check them upstream): Laravel framework v13.35.0, PHP 8.4.26, FrankenPHP v1.13.1 (Caddy v2.11.7), Swoole 6.2.3, RoadRunner 2025.1.15, oha 1.16.0, Docker 29.7.2.
• Limits: every container ran with --cpus=4 --memory=2g and 4 workers (FPM: pm = static, pm.max_children = 4).
• Load: oha, 64 concurrent connections, 20 seconds after a 5-second warm-up, from a container on the same Docker network.
• Machine: a Windows 11 laptop, Docker Desktop reporting 8 CPUs and ~8 GB. The load generator shares the machine with the server, so treat absolute numbers as low and noisy.

Laravel Octane benchmark results: FrankenPHP vs Swoole vs RoadRunner vs PHP-FPM

Three runs per cell. Each cell is the median (min–max) of those three runs. All 24 runs had a 100% success rate.

/json (returns a small JSON body):

/db (a count() plus the latest 20 rows of a 1,000-row SQLite table):