Submit benchmark results through a GitHub issue. Maintainers verify the evidence and publish approved scores through the database editor.
Prepare your evidence
Include:
- Model ID, provider, and the Phaseo page if one exists.
- Benchmark name, version, split, score, and units.
- Evaluation date, prompt format, harness, and relevant hardware or sampling settings.
- A public paper, results page, dataset, or reproducible evaluation log.
- Whether the result is provider-reported or independently measured.
Submit the issue
Request new benchmark data for a new result or benchmark. Report incorrect information to correct an existing score, including its current and proposed values.
Do not include credentials, private prompts, or customer data in public reports.
Track publication
Maintainers may ask for clarification in the issue before updating the database. Daily JSON snapshots reflect published data; editing archived benchmark JSON does not publish a score.
For help preparing a submission, use the community Discord. Last modified on September 9, 2026
