BenchHub — Build, run, and share ML benchmarks

">

Build, run, and

share ML benchmarks.

Pick a dataset, define a leaderboard, upload predictions,<br>and see how your model ranks. Open-source. No credit card.

Sign in / Sign up

See the leaderboards

How it works

Three steps from idea to ranked leaderboard.

1. Pick a dataset

Use your own data or pick one from the public catalog.<br>Images, scalars, depth maps, audio — BenchHub handles<br>the common ML data types.

2. Define a leaderboard

Pick which dataset fields submissions are scored against,<br>assign each one a role (input vs ground truth), then attach<br>the metrics and visualisations to run on every submission.

3. Submit, get ranked

Send your predictions, BenchHub scores them, and<br>your run lands on the leaderboard next to<br>everyone else's.

Sharing & storage

Every dataset and leaderboard is either public or private. Generous<br>storage on either side.

Public

50 GB

Listed on /datasets and /leaderboards for everyone

Anyone can submit predictions and appear on the board

Counted against your 50 GB public quota — covers dataset bytes, LB materialisations, and submissions

Private

10 GB

Visible only to you — not listed anywhere

Iterate without the world watching; Publish with one click when ready

Counted against your 10 GB private quota — separate from public

Active leaderboards

Most submissions in the last 30 days.

cifar100_benchmark

11

11

cifar10_benchmark

11

11

Ready to benchmark?

Sign in with GitHub, Google, or a passwordless email code.<br>We only see your name, email, and avatar.

Log in / Sign up

Confirm Action

Are you sure you want to proceed?

Cancel<br>Confirm