About

Model Follower is a nightly watch list over Hugging Face's model universe. An engine ingests the Hub's public statistics, re-checks every model on the list against the Hub API, hosted-provider listings, and the vLLM and SGLang registries, and writes the pages and feeds on this site. Every number carries its source and the time it was observed.

The site shows facts, prices, lineage, servability, and an estimate of hardware fit. It does not run benchmarks and does not publish verdicts. Where a sentence is generated, it is labeled and cites the fields it was generated from. Where data is missing, the section is absent.

Closed-weights and not-yet-released models are tracked too, on a separate frontier and closed list and an upcoming page. A nightly release radar reads router and lab API listings and new Hugging Face repos, records each model's stage (announced, API only, weights public, served by third parties) with the source it was seen at, and pairs closed models with the nearest open-weights models on the watch list by modality, context and price. That pairing is a lookup, not a quality ranking; list prices and provider descriptions are shown as observed, with their dates.

Data sources

• Hugging Face Hub (hub-stats parquet + API)
• OpenRouter
• DeepInfra
• Parasail
• vLLM registry
• SGLang
• HN Algolia
• Exa

Last generated 2026-10-25 06:54 UTC from Hub snapshot 2026-10-25. Collection uses official APIs and bulk files, conditional requests, and stays well inside published rate limits.

Methods

• Ranking: how the candidate score and section ranks are computed.
• Hardware fit: the weights-only memory estimate and classes.
• Feeds: item schema for RSS and JSON Feed consumers.

Corrections

If a fact is wrong, the source link beside it is the fastest way to check; the nightly run picks up upstream fixes automatically. Hand-curated tables (licenses, org reputation, decision-model catalog, industry pages) are reviewed files and can be corrected on request.