# How BenchLM confirms AI model releases

Model Updates publishes confirmed releases only. Every event needs a usable date and an attributable source. Expected dates, cadence forecasts, rumors, and unsourced launch claims stay out of the confirmed record.

## Confirmation rules

- A release needs a dated provider announcement, release note, model repository, documentation page, or another attributable primary listing.
- The stored date follows the source. Exact days remain exact; month-only evidence remains month precision.
- Variants released in one announcement can share one event while retaining separate model records.
- Confirmation records that a launch happened. It is not a performance, ranking, or availability judgment.

## Corrections and availability

Announcement date and broad availability can differ. The source-linked release date is preserved. When evidence changes, the shared release metadata updates the hub, timeline, provider archives, and month archives together.

Canonical page: https://benchlm.ai/model-updates/methodology