Benchmarks#

Janux measured against react 19.2, preact 10.29, solid 2.0-beta, svelte 5.56 and vue-vapor 3.6-rc across 19 multi-framework suites — client runtime, hydration, SSR, streaming and shipped bytes. The harness is a direct port of octane's benchmark infrastructure (MIT, Dominic Gannaway — see the credits below), whose correctness gates are the point: a framework that posts numbers must first prove the DOM it produced is right, keyed rows kept their identity, focus and caret survived, and every mount got its cleanup.

Everything here is reproducible from benchmarks/ in the repo: bun run bench runs the suites, node benchmarks/report.mjs renders the full position report. Numbers below are from a full run on 2026-07-29 — Apple M4 Pro, 24 GB, macOS 26.5, Node 26, Bun 1.3, Chromium 151 headless. Absolute milliseconds are a property of that machine; the ratios are the claim, and CI re-checks a committed set of ratio guards weekly.

Where Janux wins#

Resume instead of hydration. A Janux page becomes interactive by reading its serialized state snapshots and installing delegated listeners — no component re-execution, no event replay. That design shows up everywhere the hydration suites measure it:

Shipped bytes vs the React baseline. The full Janux client runtime — signals, reconciler, delegated events, resume, the agent bridge — is 22.2KB gzip; the same krausest app ships 24.2KB total against react's 60.7KB. Pages with no islands ship no JavaScript at all, which no suite here can even express.

Mass DOM work. Creating 10,000 rows: 84.9ms vs react's 131.8 (solid 15.4, svelte 27.6). Clearing them: 36.0ms vs react's 48.4. Resetting a 512-field form: 15.4ms vs react's 39.0.

Whole-app parity. On the application-shaped suites the gap to the compiled frameworks mostly disappears: mount/update/unmount lifecycle cycles at 1.00× react with exact cleanup accounting, zustand/jotai/TanStack Query integration ops between 0.83× and 1.14× (one outlier: TanStack invalidation at 1.62×), suspense error/retry/cancel recovery between 1.01× and 1.10×, a 512-subscriber selector store with zero selector calls across 20 unrelated parent re-renders.

Where Janux does not win (yet)#

Honest numbers, with the cause understood:

• Sub-millisecond keyed micro-ops. Selecting one row among 1,000 costs 7.7ms where react-with-memo pays 0.5ms and solid 0.1ms. A Janux island re-runs its whole view per state change; the reconciler now skips unchanged subtrees and moves keyed rows with minimal LIS insertions, but rebuilding the view's JSX is still O(list). Going sub-millisecond needs a fine-grained list primitive (a <For> with per-row reactive scopes) — an authoring surface change under RFC.
• SSR throughput. Rendering the 50-card news page takes 0.39ms against react's 0.09 (4.5×). Streaming end-to-end is at parity (50.7ms vs 51.3 on the staggered schedule, out-of-order boundary swaps included) because data latency dominates, but the string-emission pipeline is the next server-side target.
• Programmatic event bursts. 128 synthetic events in a tight loop cost 145ms (react 17.9): each delegated intent pays schema validation plus a microtask chain. Invisible at human interaction rates, measurable in bursts.

What the benchmarks already paid for#

Porting the harness found and fixed real framework issues before any tuning started: intent bodies now batch to one reactive flush (a krausest update was 1816ms before, 7.2ms after), keyed morph landed because the reorder suite's node-identity gate demanded it, and the SSR chunk coalescer was rewritten after the throughput suite's sustained-render loop exposed unbounded retention (~800KB pinned per render, OOM; now flat). The full optimization log, including the attempts that didn't survive, lives in benchmarks/BASELINE-2026-07.md.

Credits#

The harness, suites and rival fixtures derive from octane's benchmarks (MIT, Copyright (c) 2026 Dominic Gannaway) — runner, statistics, correctness-gate methodology and the ratio-guard CI model are theirs. The js-framework fixtures derive upstream from krausest/js-framework-benchmark (Apache-2.0). Full notices in benchmarks/LICENSE-OCTANE and CREDITS.md.