# Benchmarks

zerolith on js-framework-benchmark next to hand-written JavaScript and seven popular frameworks, measured again with every release.

## Results

The newest release, zerolith 1.5.0, measured on 2026-11-08 with medians of 15 runs on a GitHub-hosted runner, next to hand-written JavaScript and seven popular frameworks. The score is each framework's weighted geometric mean: every operation's time over hand-written JavaScript's, combined with the weights of the official results page, so hand-written JavaScript is 1.00, a score of 1.10 takes 10% longer than it over the nine operations, and lower is faster. The release before, zerolith 1.3.0, ran in the same session on the same machine, so its row compares with the newest like for like.

| Framework | Entry | Version | Score |
| --- | --- | --- | --- |
| vanilla | `vanillajs` | hand-written | 1.00 |
| zerolith, the release before | `zerolith` | 1.3.0 | 1.09 |
| Solid | `solid` | 1.9.3 | 1.10 |
| zerolith | `zerolith` | 1.5.0 | 1.12 |
| Svelte | `svelte` | 5.42.1 | 1.15 |
| Lit | `lit` | 3.2.0 | 1.22 |
| Vue | `vue` | 3.5.39 | 1.27 |
| Angular | `angular-cf-signals-nozone` | 22.2.2 | 1.35 |
| Preact | `preact-signals` | 11.0.0 + 2.3.1 | 1.36 |
| React | `react-hooks` | 19.2.0 | 1.48 |

### Every operation

Each cell is the median duration in milliseconds, with the factor against hand-written JavaScript for that operation in parentheses; below 1.00 is faster than it.

| Operation | vanilla | Solid | zerolith | Svelte | Lit | Vue | Angular | Preact | React |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| create 1,000 rows | 69.8 (1.00) | 76.9 (1.10) | 82.5 (1.18) | 78.1 (1.12) | 84.8 (1.21) | 85.3 (1.22) | 104.9 (1.50) | 108.8 (1.56) | 92.6 (1.33) |
| replace all 1,000 rows | 76.5 (1.00) | 86.8 (1.13) | 91.0 (1.19) | 94.8 (1.24) | 91.2 (1.19) | 100.1 (1.31) | 109.7 (1.43) | 116.7 (1.53) | 107.5 (1.41) |
| update every 10th row (4x slowdown) | 46.7 (1.00) | 48.9 (1.05) | 51.6 (1.10) | 50.4 (1.08) | 53.8 (1.15) | 57.7 (1.24) | 50.4 (1.08) | 52.4 (1.12) | 63.1 (1.35) |
| select a row (4x slowdown) | 8.3 (1.00) | 11.0 (1.33) | 9.3 (1.12) | 15.6 (1.88) | 17.2 (2.07) | 11.6 (1.40) | 17.3 (2.08) | 12.6 (1.52) | 18.0 (2.17) |
| swap 2 rows (4x slowdown) | 56.7 (1.00) | 62.1 (1.10) | 63.3 (1.12) | 66.3 (1.17) | 65.8 (1.16) | 65.1 (1.15) | 47.6 (0.84) | 62.2 (1.10) | 398.3 (7.02) |
| remove one row (2x slowdown) | 47.7 (1.00) | 49.5 (1.04) | 49.1 (1.03) | 48.9 (1.03) | 51.9 (1.09) | 59.3 (1.24) | 54.3 (1.14) | 50.3 (1.05) | 52.7 (1.10) |
| create 10,000 rows | 794.7 (1.00) | 865.9 (1.09) | 901.3 (1.13) | 895.3 (1.13) | 932.8 (1.17) | 970.3 (1.22) | 1077.7 (1.36) | 1122.3 (1.41) | 1126.3 (1.42) |
| append 1,000 rows to 1,000 | 88.3 (1.00) | 89.8 (1.02) | 93.1 (1.05) | 95.3 (1.08) | 97.7 (1.11) | 103.3 (1.17) | 117.4 (1.33) | 117.4 (1.33) | 113.1 (1.28) |
| clear 1,000 rows (4x slowdown) | 33.8 (1.00) | 41.8 (1.24) | 37.8 (1.12) | 41.8 (1.24) | 51.1 (1.51) | 55.2 (1.63) | 56.6 (1.67) | 57.5 (1.70) | 66.8 (1.98) |

### Memory, size, and first paint

Each value is the median of 5 passes, since the benchmark takes one sample of each a pass.

| Measure | vanilla | Solid | zerolith | Svelte | Lit | Vue | Angular | Preact | React |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| ready memory, after load (MB) | 0.56 | 0.58 | 0.73 | 0.61 | 0.70 | 0.86 | 1.11 | 0.69 | 1.18 |
| run memory, with 1,000 rows (MB) | 1.87 | 2.68 | 2.82 | 2.87 | 2.79 | 3.91 | 3.61 | 4.90 | 4.42 |
| after creating and clearing 1,000 rows 5 times (MB) | 0.64 | 0.77 | 0.97 | 0.98 | 0.88 | 1.19 | 1.69 | 8.95 | 1.98 |
| size uncompressed (kB) | 11.3 | 11.5 | 29.9 | 26.6 | 22.1 | 64.3 | 110.6 | 23.3 | 190.3 |
| size compressed (kB) | 2.5 | 4.5 | 9.0 | 9.7 | 7.3 | 23.3 | 33.8 | 8.3 | 51.4 |
| first paint (ms) | 118.4 | 121.9 | 159.2 | 139.1 | 147.0 | 216.8 | 286.2 | 139.2 | 452.0 |

### Every run

Each run's scores, the newest release first. A score compares frameworks within its own run, so a change in how the runs were taken moves every framework's score with it.

| Run | vanilla | Solid | zerolith | Svelte | Lit | Vue | Angular | Preact | React | Measured |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1.5.0, 2026-11-08 | 1.00 | 1.10 | 1.12 | 1.15 | 1.22 | 1.27 | 1.35 | 1.36 | 1.48 | medians of 15 runs on a GitHub-hosted runner |
| 1.3.0, 2026-11-07 | 1.00 | 1.12 | 1.15 | 1.17 | 1.27 | 1.29 | 1.43 | 1.38 | 1.51 | medians of 15 runs on a GitHub-hosted runner |
| 1.2.0, 2026-11-07 | 1.00 | 1.09 | 1.10 | 1.13 | 1.23 | 1.27 | 1.41 | 1.35 | 1.49 | medians of 15 runs on a GitHub-hosted runner |
| 1.1.0, 2026-11-06 | 1.00 | 1.10 | 1.10 | 1.14 | 1.24 | 1.27 | 1.45 | 1.35 | 1.49 | medians of 15 runs on a GitHub-hosted runner |
| 1.0.1, 2026-11-06 | 1.00 | 1.11 | 1.14 | 1.16 | 1.27 | 1.28 | 1.42 | 1.37 | 1.51 | medians of 15 runs on a GitHub-hosted runner |
| 1.0.0, 2026-11-05 | 1.00 | 1.13 | 1.15 | 1.17 | 1.27 | 1.26 | 1.43 | 1.37 | 1.46 | medians of 15 runs on a GitHub-hosted runner |
| 0.1.0-alpha.10, 2026-11-04 | 1.00 | 1.13 | 1.13 | 1.16 | 1.27 | 1.26 | 1.41 | 1.40 | 1.53 | medians of 15 runs on a GitHub-hosted runner |
| 0.1.0-alpha.9, 2026-11-03 | 1.00 | 1.10 | 1.13 | 1.14 | 1.24 | 1.30 | 1.41 | 1.36 | 1.46 | medians of 15 runs on a GitHub-hosted runner |
| 0.1.0-alpha.8, 2026-11-03 | 1.00 | 1.14 | 1.13 | 1.18 | 1.27 | 1.29 | 1.43 | 1.40 | 1.50 | medians of 15 runs on a GitHub-hosted runner |
| 0.1.0-alpha.8, 2026-11-02 | 1.00 | 1.07 | 1.08 | 1.10 | 1.18 | 1.19 | 1.29 | 1.27 | 1.40 | medians of 15 runs, held to one group of cores on an idle machine |
| 0.1.0-alpha.6, 2026-11-01 | 1.00 | 1.13 | 1.13 | 1.14 | 1.23 | 1.23 | 1.37 | 1.33 | 1.47 | means of 10 runs, the cores left to the system |
| 0.1.0-alpha.5, 2026-10-31 | 1.00 | 1.11 | 1.12 | 1.14 | 1.23 | 1.26 | 1.42 | 1.34 | 1.44 | means of 10 runs, the cores left to the system |

0.1.0-alpha.6, 2026-11-01: 0.1.0-alpha.7 builds the same bundle byte for byte, so this run stands for it too.

## How the runs are measured

js-framework-benchmark renders the same table of rows in every framework and times nine operations on it, from creating 1,000 rows to clearing them. Its driver clicks a button in Chrome and measures from the click to the end of the paint it causes, from Chrome's performance trace, so script, style, layout, and paint all count, and five of the operations run with the CPU slowed down, as the official runs do. Each time here is the median of a run's samples: 15 of each operation, and 25 of selecting a row.

Every framework is the keyed entry of the benchmark's repository at the commit [frameworks.json](https://github.com/molexxxx/zerolith/blob/main/benchmarks/frameworks.json) names, built from its own lockfile, and zerolith is [the entry in this repository](https://github.com/molexxxx/zerolith/tree/main/benchmarks/js-framework-benchmark), pinned to the release. The memory results are what Chrome reports for the page after an action and a forced garbage collection, and the size results are the bytes of the entry's files; the benchmark takes one sample of each a pass, so each release's run repeats them for five passes and keeps the medians.

Each release is measured once it is published, by [the benchmarks workflow](https://github.com/molexxxx/zerolith/blob/main/.github/workflows/benchmarks.yml), and the run is kept in the repository with the machine and method it used. A change of machine or method moves every framework's score, so scores compare only within one run. The earlier runs are listed under Every run, each with how it was taken.

These are this project's runs, not the official results, which the benchmark's maintainer measures and publishes at [krausest.github.io/js-framework-benchmark](https://krausest.github.io/js-framework-benchmark/); zerolith is not among them until its entry is accepted upstream. [The benchmarks README](https://github.com/molexxxx/zerolith/blob/main/benchmarks/README.md) has the full method and the steps to take a run yourself.