Models you can run yourself

A model, every quantisation somebody made of it, and what each one costs to run.

Behind2 sourcesupdated 40m agofresh within 24h
3,124things
793named by two sources or more
32conflicts
2sources

Only one source knows: GGUF quantisations 2268 · Hugging Face models 63

Covers. Every GGUF repository created since 2026-08-23, and the base model behind each one. 2,684 of those bases are named; without a Hugging Face token the run is throttled off after roughly 600 to 750 of them, so the model side of the join is partial and the scope says so on its front page. Excludes. Ollama, whose library names no Hugging Face repository to join on. Benchmarks, which have no machine-readable source.
Compared · what is held against what, and from which column of each source
PropertyHugging Face modelsGGUF quantisations
LicencecardData.licensecardData.license

Everything else the sources say is shown side by side, and not compared.

task=text-classification ✕task=text-classification ✕task=text-classification ✕licence=llama3.1 ✕

EverythingQuantised, and openly licensedmodel 1599quantisation 9105

Found

none
ThingKindLicenceDownloadsDate

Nothing here.

Facets

Kind 10704 of 10704 claims

quantisation9105
model1599

Source 10704 of 10704 claims

zetlyn/models-gguf9105
zetlyn/models-hf1599

Licence 8266 of 10704 claims

llama3.143

Sources

Hugging Face models primary

The model itself: its licence, its task and how many people fetch it.

model · failing1599

GGUF quantisations high

Which quantisations exist, which is what decides whether it runs on the machine somebody has.

quantisation · current9105