Caddy vs HAProxy vs Nginx vs Traefik: Performance Comparison 2026

Caddy, Traefik, HAProxy, Nginx and Apache solve overlapping problems — they accept HTTP requests, terminate TLS and forward traffic to applications — but they are built differently and are good in different places. Below is a comparison by what actually affects the choice, and separate analyses of the most common pairs: HAProxy or Nginx, Caddy or Traefik. Detailed posts about each server are in separate parts of the series: Nginx, HAProxy, Caddy, Traefik.

Briefly about each

• Caddy — web server and reverse proxy with automatic HTTPS by default. Apache 2.0 license, version 2.11 (2026).
• Traefik — reverse proxy that takes routes from Docker, Kubernetes and other sources. MIT license, version 3.7 (2026).
• HAProxy — specialized HTTP and TCP load balancer. GPL license, long-term branches 3.2 and 3.4.
• Nginx — web server, reverse proxy and cache, the most common “front” in front of an application. BSD license.
• Apache HTTP Server — classic modular web server with per-directory configuration (.htaccess). Apache 2.0 license, branch 2.4.

Comparison by key criteria

Until recently the certificates line was the main distinction of Caddy and Traefik from the others. Now ACME clients exist for Apache, nginx, and HAProxy too, but only Caddy and Traefik have it working out of the box without separate modules and experimental directives.

HAProxy or Nginx

Both can proxy HTTP and TCP and both are fast. The difference is what they emphasize.

HAProxy was built as a load balancer. It has advanced active health checks for servers (HTTP requests checking responses, database protocol checks), tables for rate limiting and sticky sessions, a Runtime API to take a server out of service without restarting, a detailed stats page and logs with timing for each stage of the request. It does not serve static files and is not a cache in the full sense.

Nginx is first and foremost a web server: it serves static files, works with PHP-FPM, caches responses, supports compression, URL rewriting and a rich set of modules. It can do load balancing too, but in the free version health checks are passive: a server is removed after failed user requests (max_fails), and active checks are only available in NGINX Plus.

How to choose:

• many application servers, health checks, rate limits and visibility of each request are important — HAProxy;
• a single server that needs to both serve static files and proxy the application — Nginx;
• large projects often use both: HAProxy at the edge distributes load, Nginx behind it serves static files and handles the application.

Caddy or Traefik

Both obtain certificates themselves and are well suited for small teams. The difference is the source of configuration.

Caddy is configured with a file: one or two lines per site. It’s a full web server — serves static files, works with PHP via php_fastcgi, compresses responses. Convenient on a VPS with several permanent sites and services.

Traefik is configured via container labels and Kubernetes objects. It notices when a container starts or stops and rebuilds routes. It does not serve static files — you need a separate web server in a container for those.

How to choose:

• the set of sites changes rarely, you want a simple configuration file — Caddy;
• services live in containers and constantly appear and disappear, you have Kubernetes — Traefik.

What actually affects performance

Synthetic benchmarks where one server is “twice as fast” as another say little about your load. On a typical site the bottleneck is almost always the application or the database, not the proxy. The most noticeable factors that affect the front-end speed are:

• TLS settings — session resumption, HTTP/2, modern cipher suites;
• persistent connections to application servers — without keepalive each request opens a new TCP connection;
• buffering — for streaming responses (SSE, WebSocket) it’s disabled, for regular responses it offloads the application;
• logs — writing every request to a slow disk limits throughput;
• compression and cache — serving a ready response from cache is faster than asking the application again.

If performance is critical, compare candidates with your own requests and identical TLS settings.

When to choose what

None of the five is better than the others at everything. A good setup often combines two: a load balancer at the edge and a web server behind it, or Traefik for containers and nginx for static files inside them.