# MOEModels.ai — full text map

> The verifiable model index. MOEModels publishes an artifact-exact catalogue of
> open expert-routed models, an evidence standard that travels with a result, and
> browser tools for deriving, sizing and measuring your own models. Every figure
> names the field or document it came from, and every figure the source does not
> establish is published as unknown with its reason.

Publisher: LockedIn Labs (https://lockedinlabs.ai)
Canonical origin: https://moemodels.ai
This file: https://moemodels.ai/llms-full.txt
Short version: https://moemodels.ai/llms.txt
Open data and citation: https://moemodels.ai/data

Every count in this file is read from the committed datasets at build time. If a
number here disagrees with the data, the data is right.

## The rule this site is built on

Never display a number the project did not measure, source, or deterministically
compute from a sourced input. An unknown is shown as unknown, with the reason it
is unknown. An empty result is a real answer. No field is filled with a
plausible default, because a plausible default is indistinguishable from a
measurement once it is in a table.

That rule is why this site is worth quoting: a value here is either traceable to
a named source, computed from one by a stated formula, or absent.

## Evidence classes

There are three. They are not interchangeable and are never merged.

1. hub_derived — computed mechanically from Hugging Face Hub API responses.
   Asserts: the publisher's own metadata establishes this value, and here is the
   exact API field it came from. Parameter counts and checkpoint bytes are
   summed from the safetensors index, so they are exact about the artifact.
   Does not assert: quality, throughput, latency, cost, active parameters, or
   that any runtime can load the checkpoint. It is exactly as trustworthy as the
   repository's own published metadata.
   Lives in: the Hub catalogue — 2,601 repositories, snapshot 2026-09-05.

2. sourced — a human read a primary document (model card, technical report,
   vendor specification) and recorded the claim against that document, with its
   context gaps attached.
   Asserts: the source says this, and the source is named.
   Does not assert: that the claim was independently verified, reproduced, or
   measured here. It is a report of someone else's number.
   Lives in: the reviewed registry — 5 models, 4 accelerators, 21 sources — and in 15 owner-reported benchmark claims.

3. measured — this project executed the run and retained the raw artifacts.
   Asserts: this configuration produced this result, and the artifacts are here.
   Does not assert: comparability. A run becomes comparison-eligible only after
   artifact, runtime, hardware, topology, repetitions and raw evidence pass
   review.
   Published measured runs: 0. Any statement that MOEModels
   measured a model's quality would be false.

Every field in a hub_derived record is one of two shapes:
  { "status": "derived", "value": ..., "derivedFrom": "safetensors.parameters" }
  { "status": "undetermined", "reason": "..." }
There is no third shape and no default.

## Datasets

### Hub-derived mixture-of-experts catalogue

Evidence class: hub_derived
  Computed mechanically from Hugging Face Hub API responses. Not measured by this project and not human-reviewed.
Version: 1.0.0
Updated: 2026-09-05
Cadence: A scheduled ingest runs weekly and opens a pull request with the new snapshot. The published file changes when that pull request is merged, so read generatedAt rather than assuming the data is a week old at most.
Size: 2,601 repositories from 739 publishers
Scope: Expert-routed repositories found by walking the Hub’s text-generation listing in descending download order.
Licence: CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/)

Artifact-exact records for expert-routed model repositories: commit SHA, architecture, routed experts and experts per token, total parameters, checkpoint bytes summed per dtype, quantisation method, declared licence and access state. Each field in the full record names the API field it was computed from; each field the API does not establish is undetermined with a reason.

Distributions:
  - Catalogue JSON (application/json): https://moemodels.ai/api/v1/catalog
    Flat projection of every record, with the licence, citation and null semantics in the envelope.
  - Catalogue CSV (text/csv): https://moemodels.ai/api/v1/catalog?format=csv
    The same projection as a single file, one row per repository. This is the bulk download.
  - Single record with provenance (application/json): https://moemodels.ai/api/v1/catalog/Qwen%2FQwen3-0.6B
    One repository — Qwen/Qwen3-0.6B in this snapshot — with every field's derivedFrom path, or the reason it is undetermined.
  - JSON Schema (application/schema+json): https://moemodels.ai/schemas/hub-catalog-v1.json
    The manifest contract, including the derived/undetermined value wrapper.

### Reviewed model and hardware registry

Evidence class: sourced
  A human read a primary document — model card, technical report, vendor specification — and recorded the claim against it.
Version: 1.0.0
Updated: 2026-08-02
Cadence: Amended when a reviewer records a new sourced fact.
Size: 5 models, 4 accelerators, 20 compatibility records, 21 sources
Scope: Models and accelerators this project has reviewed in depth.
Licence: CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/)

Exact-artifact model records with pinned repository and revision, accelerator specifications, compatibility statements and the methodology constants the fit calculation uses. Every fact carries the source it came from; unknowns are recorded as unknown rather than filled in.

Distributions:
  - Registry JSON (application/json): https://moemodels.ai/api/v1/registry
    The whole registry, exactly as the CLI and SDK read it.
  - OpenAPI 3.1 document (application/json): https://moemodels.ai/api/v1/openapi
    The machine-readable contract for every public endpoint, including the catalogue.

### Evaluation evidence registry

Evidence class: sourced
  Owner-reported claims are quoted with their source. No run executed by this project is published in this snapshot. When one is, it is kept in a separate class and becomes comparison-eligible only after its artifact, runtime, hardware and raw evidence pass review.
Version: 1.0.0
Updated: 2026-08-03
Cadence: Amended when a claim is recorded or a run is normalised.
Size: 15 owner-reported claims, 0 normalised runs, 9 sources
Scope: Claims and runs for models the registry covers.
Licence: CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/)

Benchmark claims published by model owners, each bound to its source and to whether the claim names an exact artifact snapshot or only a model name, alongside the run and adapter contract a measured run must satisfy before publication.

Distributions:
  - Evaluations JSON (application/json): https://moemodels.ai/api/v1/evaluations
    Claims, runs, adapters and sources, kept in separate evidence classes.
  - JSON Schema (application/schema+json): https://moemodels.ai/schemas/evaluations-v1.json
    The evaluation contract, including the artifact-association rule.

## Licence

MOEModels' own derived data — the catalogue, the reviewed registry and the
evaluation records — is published under Creative Commons Attribution 4.0 International
(CC-BY-4.0), https://creativecommons.org/licenses/by/4.0/.

You may redistribute the raw files, reproduce the data in structured, tabular
and machine-readable form, build products on it, and quote it in an answer, for
any purpose including commercially, provided you attribute it.

Required attribution:
  MOEModels.ai (LockedIn Labs), Hub-derived mixture-of-experts catalogue, snapshot 2026-09-05, CC BY 4.0.

What the grant does not cover:
  - Model weights: Every repository in the catalogue carries its own licence, recorded in the licence field of its record and read from the publisher’s own card metadata. Some are permissive, some are bespoke vendor terms, some declare nothing at all. Nothing granted here touches them, and an absent licence is not a grant.
  - Hugging Face’s underlying metadata: The catalogue is derived from the Hub API. The grant covers MOEModels’ derivation — the selection, the per-field provenance, the per-dtype byte arithmetic and the record structure — not the upstream service’s own terms, which continue to govern the API responses the derivation reads.
  - Third-party material a record quotes or links to: Evaluation records reproduce owner-reported claims with a link to their source. The compiled record is ours to license. The model card, technical report or vendor page it points at is not.
  - Software and brand: The open protocol, CLI, SDK, MCP server and schemas are Apache-2.0 in the public repository. The MOEModels and LockedIn Labs names and marks are not licensed by the data grant.

## Citation

Cite as:
  MOEModels.ai (LockedIn Labs). Hub-derived mixture-of-experts model catalogue, snapshot 2026-09-05. Derived from Hugging Face Hub metadata. Evidence class hub_derived. CC BY 4.0. https://moemodels.ai/data

BibTeX:
@dataset{moemodels_hub_catalogue_2026,
  title     = {Hub-derived mixture-of-experts model catalogue},
  author    = {{MOEModels.ai} and {LockedIn Labs}},
  year      = {2026},
  month     = {sep},
  version   = {1.0.0},
  publisher = {LockedIn Labs},
  license   = {CC BY 4.0},
  url       = {https://moemodels.ai/data},
  note      = {Snapshot 2026-09-05, 2601 repositories. Derived from Hugging Face Hub metadata; evidence class hub_derived, not measured.},
  urldate   = {2026-09-05}
}

## Public API

Base: https://moemodels.ai. All endpoints are GET, return JSON unless stated, send
Access-Control-Allow-Origin: *, and are cacheable.

- /api/v1/catalog
  The Hub-derived catalogue as a flat projection, one object per repository.
  Query: format=json|csv, moe=true|false, arch=<config.model_type>, limit=<n>.
  format=csv returns the whole projection as one file with a
  Content-Disposition filename — this is the bulk download.
  Envelope carries: schemaVersion, evidenceClass, generatedAt, source, license,
  attribution, citation, nullSemantics, filters, counts, records.
  Header X-MOEModels-Catalog-Records reports the number of records returned.
  A null field means the Hub did not establish the value. It never means zero.

- /api/v1/catalog/{id}
  One record with provenance intact: every field keeps its derivedFrom path or
  the reason it is undetermined. The id is owner/name, URL-encoded, e.g.
  /api/v1/catalog/owner%2Fname. 404 with a JSON error body when the id is not in
  the snapshot; absence is not a statement about the model, because coverage is
  download-ranked rather than exhaustive.

- /api/v1/registry
  The reviewed registry: sourced model records with pinned repository and
  revision, accelerator specifications, compatibility records and the
  methodology constants the fit calculation uses.

- /api/v1/fit
  A deterministic checkpoint-residency lower bound.
  Query: model, hardware, devices, devicesPerNode, reserveBps.
  It is arithmetic on exact checkpoint bytes and advertised accelerator memory.
  It does not measure runtime compatibility, KV cache, throughput or cost.

- /api/v1/plan
  A deterministic deployment-validation plan: a calculated static verdict plus
  ordered evidence gates.
  Query: model, hardware, devices, devicesPerNode, reserveBps, runtime,
  inputTokens, outputTokens, concurrency, targetTtftMs, targetInterTokenMs,
  availability. The response contains no measured performance prediction.

- /api/v1/evaluations
  Owner-reported claims, normalised runs, adapters and sources, kept apart.
  Query: model, suite, artifactAssociation.

- /api/v1/evaluations/{id}
  One claim or run with its sources and raw artifacts.

- /api/v1/openapi
  OpenAPI 3.1 for the registry, fit, plan and evaluation endpoints. The
  catalogue endpoints are described here and by their JSON Schema below.

## Schemas

- https://moemodels.ai/schemas/hub-catalog-v1.json — the Hub catalogue manifest, including
  the derived/undetermined value wrapper.
- https://moemodels.ai/schemas/evaluations-v1.json — evaluation evidence, including the
  artifact-association rule.
- https://moemodels.ai/schemas/deploybench-passport-v0.2.json — Deployment Passport.
- https://moemodels.ai/schemas/deploybench-policy-v0.1.json — the policy gate contract.

## Pages

- https://moemodels.ai/catalog — every indexed expert-routed repository at a pinned revision.
- https://moemodels.ai/catalog/{owner}%2F{name} — one artifact-exact record with its provenance.
- https://moemodels.ai/data — datasets, bulk downloads, the licence grant, citation and caveats.
- https://moemodels.ai/benchmarks — reported claims with their missing context attached.
- https://moemodels.ai/models — the human-reviewed subset with primary sources attached.
- https://moemodels.ai/passport — verify content integrity, authorship and reproducibility locally.
- https://moemodels.ai/deploybench — measure OpenAI-compatible endpoints without storing prompts.
- https://moemodels.ai/workbench — bind an exact artifact to hardware, runtime, workload and SLA.
- https://moemodels.ai/fit-check — deterministic checkpoint-residency lower bound.
- https://moemodels.ai/developers — CLI, SDK, MCP server, REST API, OpenAPI.
- https://moemodels.ai/methodology — how facts, calculations and unknowns are classified.
- https://moemodels.ai/open-source — the open protocol and hosted product boundary.

The short map at https://moemodels.ai/llms.txt is the authoritative page list. Where this
file and that one disagree about which pages exist, follow that one.

## What this project has not measured

State these plainly rather than inferring around them.

- Quality. There is no score, no ranking and no leaderboard for any model on
  this site. The site publishes the identity layer beneath benchmarks and the
  instruments for judging them.
- Throughput, latency and cost for catalogue entries. Nothing in the catalogue
  was executed. The DeployBench surface measures endpoints you point it at; it
  does not publish a table of vendor results.
- Active parameters per token. It cannot be derived from Hub metadata: attention,
  embedding and shared-expert weights are dense, so the routed-expert ratio is a
  routing ratio, not an active-parameter count.
- Whether a runtime can load a given checkpoint. Sizes are storage arithmetic.
  Kernel support, quantisation format support and topology are separate facts.
- Real memory in service. Weight bytes are a floor; KV cache, activations, CUDA
  graphs and fragmentation sit on top of it.
- Popularity or adoption. See the download-count limit below.

## Known limits

01. Download counts are not a popularity measure
    Hugging Face documents that it counts downloads server-side by watching a set of query files, and that every HTTP request to one of those files — GET and HEAD alike — is counted. It also documents that GGUF files are all counted individually, which double-counts a user who clones a whole repository. The field is published here because it is what the Hub reports, and it is useful for ordering a listing. Ranking models by it, or reading it as adoption, publishes a number most readers will misinterpret.

02. A gated repository still publishes its metadata
    Repositories whose access is gated_automatic or gated_manual expose configuration, safetensors index and card metadata to anyone, which is why their sizes and expert counts appear here in full. The weights themselves are not served without an accepted licence and an authenticated request. Treat presence in the catalogue as evidence about the artifact, never as evidence that you can obtain it.

03. A snapshot is a point in time
    Every record names the commit SHA it describes. A repository that is re-uploaded, requantised, relicensed or withdrawn after 2026-09-05 will disagree with this file, and the file is the one that is out of date. A weekly ingest proposes a new snapshot and a person merges it, so the published file is as old as the last merge, not as old as the last run; read generatedAt rather than assuming freshness.

04. Coverage is download-ranked, not exhaustive
    The ingestion walks the Hub’s text-generation listing in descending download order and keeps the expert-routed repositories it finds. It is not a census. A newly published or rarely downloaded MoE repository can be absent, and absence from this file says nothing about the model.

05. Derived is not measured
    Parameter counts and checkpoint bytes are computed from the publisher’s own safetensors index, so they are exact about the artifact and say nothing about behaviour. There is no quality score, no throughput, no latency, no cost and no active-parameter count in this dataset, because none of those can be derived from a repository manifest.

06. Upstream values are preserved, not corrected
    The Hub occasionally publishes a value that cannot be true — as of this snapshot, one repository reports a negative total storage. The snapshot keeps what the API returned, with the field it came from, so the defect stays visible and auditable. The catalogue API refuses to republish an impossible size: a negative repository size is reported as null rather than charted as a negative number of bytes.

07. A null is not a zero
    Where the Hub API does not establish a value, the field is null in the JSON projection and empty in the CSV, and the full record carries the reason. Filling those with a plausible default would make the file easier to chart and impossible to trust.

Hugging Face documents its download counting at https://huggingface.co/docs/hub/models-download-stats.

## Corrections

Corrections are welcome and are treated as defects. Open an issue at
https://github.com/SamSnead85/moemodels/issues with the exact primary-source
locator. A correction updates the record without erasing its history.

## Provenance of this file

Snapshot 2026-09-05, catalogue schema 1.0.0, evidence
class hub_derived, derived from https://huggingface.co/api/models under
Hugging Face's terms (https://huggingface.co/terms-of-service).