Use SelfBench
Publish Results
Release aggregate benchmark results for a public repository.
After evaluating approved tasks, open your repository’s Releases page. Preview the available model, harness, and reasoning settings, choose the settings to include, and publish a release. The repository must be public on GitHub and the selected settings must share at least one approved task.
A release displays aggregate accuracy and cost on selfbench.dev. A repository can have multiple publisher workspaces; each line shows its newest release that has not been withdrawn. You can withdraw a release, and the line falls back to its prior release.
Per-task trial results are not part of the public release. Publishing task files is a separate optional choice in the release flow; review that choice before publishing if your benchmark tasks should remain private. Anyone can read the released aggregates through the Public Results API.