Open data

The whole catalog, as an open dataset. Every night we publish the public catalog metadata as a single versioned JSON file with a checksum — so the catalog never dies and anyone can mirror, analyse or re-host it.

Download

Download latest dump (JSON) SHA-256 checksum

Schema version: 1
Generated: 2026-07-07T04:21:02Z
Size: 151,627 bytes
Models: 96

SHA-256

00d16642f07ed61d7433a8e9c04b7b004cfd44c56e839b82fdad4dada1088c93

Verify with: sha256sum -c llmt-open-data.json.sha256 (download the checksum file alongside the JSON).

Stable URLs — https://llmtorrents.com/open-data/latest.json always resolves to the newest dump.

What is in it

A single JSON document with a versioned envelope:

{
  "schema_version": "1",
  "generated_at": "…Z",
  "source": "…",
  "license": "CC0-1.0",
  "counts": { "models": …, "collections": …, "bundles": … },
  "data": {
    "models":      [ { slug, name, publisher, architecture, license,
                       quants: [ { format, quant_level, download } ] } ],
    "collections": [ { slug, name, models: [slug, …] } ],
    "bundles":     [ { slug, name, models: [slug, …], download } ]
  }
}

Schema notes

Downloads. Each quant has a download block. For freely-licensed models it carries the magnet, BitTorrent v1/v2 info hashes, webseeds and per-file paths + SHA-256 hashes.
Gated models. Models under a click-through license appear as metadata only — their download block is {"state":"license_acceptance_required"} with a pointer to the model page. No magnet, hash or file data is included.
Only published content. Only torrents that have passed our pre-publish gate are referenced; anything else shows {"state":"not_available"}.
Provable claims only. The hash verdict is the raw stored value (hash_matched / mismatch / no_authoritative_source) — never a synthesised "safe".
No personal data. This is catalog metadata only. There are no users, emails, downloads-per-user or any other personal information.
Deterministic. Entities are stably ordered, so day-to-day diffs are meaningful. The only field that changes when the catalog is unchanged is generated_at.

Licence & reuse

The dataset is published under CC0-1.0 (public domain dedication). Mirror it, re-host it, build on it. It is also available programmatically through the public JSON API.