# Share a hyperfine Benchmark — Numbers With Error Bars, as a Link

Canonical: https://commareports.com/share-hyperfine-benchmark
Published: 2026-09-15

> hyperfine exports JSON and Markdown, and ships plotting scripts. Render the comparison, publish it to Comma, and the 'it's faster now' claim comes with a URL.

# Share a hyperfine benchmark

hyperfine is the tool that makes "it's faster now" a checkable claim: warmup
runs, statistical outlier detection, and a standard deviation next to every
mean. Then the result gets pasted into a pull request as a screenshot of a
terminal, and the standard deviation — the number that decides whether the
change is real — is the first thing nobody reads.

## Markdown → HTML

```bash
hyperfine --warmup 3 \
  './old-binary --input big.json' \
  './new-binary --input big.json' \
  --export-markdown bench.md

pandoc bench.md -s -o bench.html
```

The Markdown table hyperfine emits is already the comparison, with relative
speedup and the ± spread. `pandoc -s` makes it a standalone page. See
[sharing pandoc HTML](/share-pandoc-html).

## JSON → a plot

```bash
hyperfine --warmup 3 -m 30 \
  './old-binary' './new-binary' \
  --export-json bench.json

python scripts/plot_whisker.py bench.json -o bench.png
```

The plotting scripts in hyperfine's repository (`plot_whisker.py`,
`plot_histogram.py`) read the JSON. If you want something interactive rather
than a PNG, feed the same JSON into a
[Plotly](/share-plotly-html) or [Vega-Lite](/share-vega-lite-chart) figure and
publish that — the per-run timings are all in the file.

## Publish it

Drag `bench.html` into [the app](https://commareports.com/), or from CI:

```bash
curl -fsS -X PATCH "https://commareports.com/api/v1/reports/$REPORT_ID" \
  -H "Authorization: Bearer $COMMA_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --rawfile html bench.html \
        --arg title "Benchmark — $GITHUB_REF_NAME" '{title: $title, html: $html}')"
```

## Benchmarks want a schedule, not a commit hook

CI runners are shared, throttled and noisy; per-commit benchmarks mostly
measure the neighbours. A nightly run on a consistent machine PATCHing one
report id gives you a revision history where a real regression stands out
against the noise floor instead of drowning in it.

- **A [routine](/features/routines/scheduled-html-reports)** does the
  re-running and the publishing.
- **Anchored threads** on a specific row when the number is disputed. See
  [commenting on HTML](/comment-on-html).

## Limits

- **Entry HTML: 5 MB.** Assets: 25 MB per file, 250 MB and 500 files total.
- **60 requests/minute per token.**

## Try it

Comma is free — unlimited reports, unlimited commenters, unlimited revision
history.

**[Publish a benchmark →](https://commareports.com/)**

### Related

- [Share a benchmark report](/share-benchmark-report) · [Share a Criterion benchmark report](/share-criterion-benchmark-report)
- [Share a JMH benchmark report](/share-jmh-benchmark-report) · [Share a pprof profile](/share-pprof-profile)
- [Share a flame graph](/share-flamegraph) · [Publish from CI](/docs/ci)