Open LLM Distribution Leaderboard

What counts Public Hugging Face model repos above the minimum download threshold are scanned, then filtered for LLM-ish text, coding, multimodal, and special architecture signals.

Models Related repos are grouped by normalized model name and explicit base-model metadata when available. The main row uses the most-downloaded repo in that group.

Metric scope Grouped metrics sum downloads, all-time downloads, and likes across related lineage repos, including package, quantized, and component repos when metadata or slug evidence links them safely. Representative metrics use only the main repo shown in the Model column.

Default sort Rows are sorted by average downloads per month unless you choose likes per month. Sorting follows the selected metric scope, and visible rank is recalculated after filters.

Rates avg downloads/month = all-time downloads / max(model age, 1 month)
avg likes/month = total likes / max(model age, 1 month)

Coverage Discovery scans configured creators, router models, and popular artifact searches. Valid popular artifacts without a safe grouping target stay as standalone rows. The snapshot omits repos below the configured download floor, currently 100K.

Age filter The default view hides models older than one year. Use Off to include older models from the generated snapshot.

Snapshots The leaderboard updates daily. Each refresh writes a new latest view and keeps timestamped historical snapshots in the bucket. Ages are calculated against the snapshot generation time.

Download counts Hugging Face download counts are public, deduplicated counts over tracked model files, meant to approximate model pulls rather than raw requests for every split model file. The recent downloads field is a recent-window count; downloadsAllTime is the same style of count over all time. This leaderboard uses downloadsAllTime divided by model age for Avg Downloads / Month, while recent downloads appear only in metric tooltips.