# aimodelscomparison.com

> Independent catalog and comparison engine for the ~1,000 most-used open AI models on Hugging Face. Unique data: daily download snapshots since 28 Jul 2026 (day-by-day history Hugging Face itself does not expose), estimated VRAM per quantization, licence commercial-use classification, and cached real prompt outputs run identically across models.

Data is refreshed nightly (03:00–04:45 UTC) from the public Hugging Face Hub API. All numbers are reproducible; methods are documented at http://aimodelscomparison.com/methodology.

## Datasets

- Daily snapshots: trailing-30d downloads, all-time downloads and likes per model per day
- Model specs: parameters, context length, architecture, licence (+ commercial-use class: yes/conditional/no), gated status, file inventory with sizes
- VRAM estimates: file_size × 1.1 + kv_cache(8K ctx) + 0.5 GB overhead, per quantization
- Versus pairs: canonical comparisons within one task, scored by shared popularity and size proximity

## Key pages

- http://aimodelscomparison.com/trending: fastest-growing models by 7-day download growth
- http://aimodelscomparison.com/models: filterable catalog (task, hardware, licence, size)
- http://aimodelscomparison.com/model/[slug]: one model — trend, VRAM table, specs, family
- http://aimodelscomparison.com/compare/[a]-vs-[b]: side-by-side specs, VRAM, trends, real outputs
- http://aimodelscomparison.com/best: data-ranked collections

## Machine access (no key required, 60 req/min)

- GET http://aimodelscomparison.com/api/v1/models/[slug] — full model record as JSON
- GET http://aimodelscomparison.com/api/v1/trending?days=7&task=[pipeline_tag] — growth ranking as JSON

Attribution appreciated: link the model or comparison page you cite.