Benchmarks

Local oha runs on the same machine, same window, same runtime version. Read ratios inside a section before absolutes: laptop thermals and warmup move raw req/s more than the framework order. Reproduce with bun run bench:ssr and bun run bench:http.

Full-stack SSR - Nifra vs the meta-frameworks

One section per UI framework, each with two explicitly-labelled tables: A - dynamic SSR (the page rendered on every request, no caching) and B - cacheable (the same page from each framework's SSG/ISR mode). Meta-frameworks run on Node through their own production server. Nifra shows Bun and Node rows (@nifrajs/node) so the Node rows compare apples-to-apples. The headline multipliers below are Table A only.

Backend - HTTP throughput across runtimes

Nifra is also a standalone API framework. Two core workloads - a path-param GET and a validated POST - each runtime through Nifra's real adapter, next to that runtime's raw handler and the popular libraries. Rows are ordered by the geometric mean of the workloads, so no single column decides the ranking.

Backend - the same routes with a real route's work

The tables above are the shape every framework publishes: a bare route doing nothing but routing. This is the same GET and POST after the route gains what an actual API route has - security headers, CORS, a request-id hook, bearer auth through a derive, and a cookie read. That combination disqualifies the route from every one of Nifra's fused fast lanes, so these are the pessimistic numbers, not the flattering ones. The third column adds one body-observing middleware to the same GET.

Reproduce locally with bun run bench:http:update, bun run bench:http:realworld:update, and bun run bench:ssr. Same-run ratios matter more than absolute req/s.