--- title: Benchmarks description: One app built three ways — rshono, Next.js and TanStack Start — measured on payload bytes, build time, cold start and install size. --- _Generated from `packages/benchmarks` on 2026-09-03 (build, payload, coldstart, load, devstart). One run, one machine. Regenerate with `pnpm --filter @rshono/benchmarks bench` and `pnpm --filter @rshono/benchmarks site:publish`._ ## Environment | Property | Value | | -------- | ------------------------------------------------------ | | Machine | Apple M1 · 8 cores · 16 GB | | Platform | darwin 25.5.0 arm64 | | Node | v22.22.2 | | CI | no — laptop numbers, treat spreads under ~10% as noise | ### Versions React version skew across the three is unavoidable — Next vendors its own copy — so it is reported rather than hidden. A render-path difference of a few percent is more likely this than the framework. | Package | rshono | Next.js | TanStack Start | | ------------------------- | ----------- | ------- | -------------- | | `react` | 19.2.8 | 19.2.8 | 19.2.8 | | `react-dom` | 19.2.8 | 19.2.8 | 19.2.8 | | `@rshono/core` | 1.0.0-rc.20 | — | — | | `hono` | 4.13.5 | — | — | | `@rspack/core` | 2.2.2 | — | — | | `react-server-dom-rspack` | 0.1.0 | — | — | | `next` | — | 16.2.12 | — | | `@tanstack/react-start` | — | — | 1.168.33 | | `@tanstack/react-router` | — | — | 1.170.18 | | `vite` | — | — | 7.3.6 | ## Initial-load payload Brotli-compressed bytes the browser is committed to fetching before the route is interactive: the document, the inline flight payload, and every statically referenced script and stylesheet. Compression is applied by the harness, identically for all three. ### `/` — prerendered | Metric | rshono | Next.js | TanStack Start | | ------------------- | ---------------- | ------------------ | ----------------- | | Document (br) | 871 B | 1.7 kB | 1.1 kB | | Inline script (raw) | 2.3 kB | 5.7 kB | 1.3 kB | | External JS (br) | 58.9 kB · 1 file | 160.0 kB · 9 files | 97.2 kB · 2 files | | CSS (br) | 657 B · 1 | 615 B · 1 | 611 B · 1 | | **Total (br)** | **60.4 kB** | **162.3 kB** | **98.8 kB** | | Total (raw) | 223.6 kB | 627.8 kB | 356.6 kB | | Requests | 3 | 11 | 4 | | Spec checks | ✓ | ✓ | ✓ | ### `/ssr` — dynamic | Metric | rshono | Next.js | TanStack Start | | ------------------- | ---------------- | ------------------ | ----------------- | | Document (br) | 4.3 kB | 5.2 kB | 5.3 kB | | Inline script (raw) | 35.1 kB | 38.9 kB | 37.4 kB | | External JS (br) | 58.9 kB · 1 file | 160.0 kB · 9 files | 97.0 kB · 2 files | | CSS (br) | 657 B · 1 | 615 B · 1 | 611 B · 1 | | **Total (br)** | **63.9 kB** | **165.8 kB** | **102.9 kB** | | Total (raw) | 268.8 kB | 673.4 kB | 404.3 kB | | Requests | 3 | 11 | 4 | | Spec checks | ✓ | ✓ | ✓ | ### `/interactive` — dynamic | Metric | rshono | Next.js | TanStack Start | | ------------------- | ----------------- | ------------------- | ----------------- | | Document (br) | 2.9 kB | 3.7 kB | 3.9 kB | | Inline script (raw) | 13.4 kB | 17.6 kB | 15.7 kB | | External JS (br) | 60.0 kB · 4 files | 160.9 kB · 10 files | 98.5 kB · 4 files | | CSS (br) | 657 B · 1 | 615 B · 1 | 611 B · 1 | | **Total (br)** | **63.5 kB** | **165.2 kB** | **102.9 kB** | | Total (raw) | 239.6 kB | 644.7 kB | 376.6 kB | | Requests | 6 | 12 | 6 | | Spec checks | ✓ | ✓ | ✓ | ### `/api/health` — json | Metric | rshono | Next.js | TanStack Start | | ------------------- | ------------- | ------------- | -------------- | | Document (br) | 32 B | 32 B | 32 B | | Inline script (raw) | 0 B | 0 B | 0 B | | External JS (br) | 0 B · 0 files | 0 B · 0 files | 0 B · 0 files | | CSS (br) | 0 B · 0 | 0 B · 0 | 0 B · 0 | | **Total (br)** | **32 B** | **32 B** | **32 B** | | Total (raw) | 28 B | 28 B | 28 B | | Requests | 1 | 1 | 1 | | Spec checks | ✓ | ✓ | ✓ | ## Build Median of 3 trials. Cold clears the framework's cache directory first; warm keeps it and touches one source file the interactive route imports. | Metric | rshono | Next.js | TanStack Start | | ------------- | --------- | ---------- | -------------- | | Cold build | 533ms ±3% | 3.50s ±15% | 3.02s ±1% | | Warm rebuild | 537ms ±2% | 3.37s ±9% | 3.00s ±0% | | Build output | 2.32 MB | 5.92 MB | 1.57 MB | | Output files | 20 | 198 | 30 | | Server bundle | 381.9 kB | — | 804.3 kB | ## Cold start Process spawn to first answered request, fresh process each trial. Not a real serverless cold start — no container, no network — it isolates the JavaScript the framework has to parse and run before it can respond. | Metric | rshono | Next.js | TanStack Start | | ---------------------- | --------- | --------- | -------------- | | Spawn → first response | 216ms ±1% | 330ms ±1% | 528ms ±0% | | Server bundle | 381.9 kB | — | 804.3 kB | ## Throughput 32 connections, 8s per route after a 2s warmup, driven by the harness's own Node load generator. **Read this as a floor check, not a headline.** All three render through the same React and stream through the same react-dom, so a large gap would mean an HTTP layer is pathological rather than that one framework renders faster. The in-process driver is identically handicapping for all three, and its absolute rps is a lower bound. `/api/health` is the informative row: no React on the path, so it is router and response construction alone. All three put React server components on the request path for `/ssr` and `/interactive` (APP_SPEC.md rule 8), so those two rows compare implementations of one architecture. They are not a perfect match: rshono and Next encode and decode the whole document, TanStack Start only the route body its RSC helpers wrap — its shell and nav stay on the cheaper non-RSC path. The flight round trip dominates both rows; on `/ssr` it is roughly 85% of the request. ### `/` | Metric | rshono | Next.js | TanStack Start | | ------------ | ------ | ------- | -------------- | | Requests/sec | 30,922 | 4,932 | 3,971 | | p50 | 0.94ms | 5.43ms | 6.83ms | | p99 | 2.15ms | 19ms | 22ms | | Errors | 0 | 0 | 0 | ### `/ssr` | Metric | rshono | Next.js | TanStack Start | | ------------ | ------ | ------- | -------------- | | Requests/sec | 566 | 276 | 253 | | p50 | 54ms | 115ms | 120ms | | p99 | 105ms | 142ms | 238ms | | Errors | 0 | 0 | 0 | ### `/interactive` | Metric | rshono | Next.js | TanStack Start | | ------------ | ------ | ------- | -------------- | | Requests/sec | 1,995 | 542 | 933 | | p50 | 15ms | 54ms | 31ms | | p99 | 25ms | 143ms | 62ms | | Errors | 0 | 0 | 0 | ### `/api/health` | Metric | rshono | Next.js | TanStack Start | | ------------ | ------ | ------- | -------------- | | Requests/sec | 25,933 | 3,361 | 12,037 | | p50 | 1.15ms | 7.14ms | 2.20ms | | p99 | 2.44ms | 42ms | 7.09ms | | Errors | 0 | 0 | 0 | ### Memory Resident memory of the whole process tree, and of the single largest process in it — which is the server itself in all three. The tree total carries whatever `npm run start` left running and double-counts pages the processes share, so the **server** row is the one to compare. **None of these are retained-memory figures.** RSS is a high-water mark that includes garbage V8 has not collected yet, and V8 sizes the old generation against the *allocation rate* — so under a fixed-duration load the fastest server churns the most and grows the largest heap. On this app a forced GC returned 362 MB of the 472 MB an uncapped `/api/health` run reported. All three are therefore given the same old-space budget (`--max-old-space-size=256`), which is what makes the rows comparable; the per-route sequence is there so a plateau is distinguishable from a climb. Retention per request measured on the rshono app, after a full GC, was under 20 B — a leak is not what these numbers show. | Metric | rshono | Next.js | TanStack Start | | ----------------------- | --------------------------------------------- | --------------------------------------------- | --------------------------------------------- | | RSS idle — tree | 155.95 MB (2 procs) | 160.86 MB (2 procs) | 242.75 MB (3 procs) | | RSS idle — server | 90.63 MB | 95.42 MB | 160.08 MB | | RSS after load — tree | 235.61 MB (2 procs) | 259.84 MB (2 procs) | 299.36 MB (3 procs) | | RSS after load — server | 173.36 MB | 197.70 MB | 219.58 MB | | RSS per route — server | 127.36 MB → 166.80 MB → 173.25 MB → 173.36 MB | 141.52 MB → 192.38 MB → 196.66 MB → 197.70 MB | 191.69 MB → 213.88 MB → 219.48 MB → 219.58 MB | | Requests served | 475,462 | 73,040 | 137,668 | | Churn per 1k requests | 178.2 kB | 1.40 MB | 442.6 kB | ## Dev server startup `dev` command to a served `/interactive` — which every one of these compiles lazily, so it includes compiling a route with three client components rather than just binding a socket. Cold clears the dev cache first. HMR round-trip is the other number worth having here and is not measured: it needs a browser driving the page to assert the patch arrived. | Metric | rshono | Next.js | TanStack Start | | -------------- | ---------- | ---------- | -------------- | | Cold dev start | 508ms ±12% | 1.53s ±10% | 2.33s ±6% | | Warm dev start | 506ms ±2% | 1.51s ±1% | 2.21s ±2% | ## Footprint A production-only install (`--omit=dev`) into a throwaway directory, and the application code the spec took to express. | Metric | rshono | Next.js | TanStack Start | | ------------------- | -------- | --------- | -------------- | | Prod install size | 72.82 MB | 291.09 MB | 57.47 MB | | Packages installed | 16 | 22 | 101 | | Direct dependencies | 4 | 3 | 4 | | App source files | 16 | 16 | 18 | | App source lines | 462 | 458 | 670 |