Inferit model catalog

The models the Inferit API knows, as open data: ids, names, aliases, context windows and licence classes.

The same rows GET /v1/models serves, without the prices and seller data the API serves live.

Status: Inferit is a testnet demo on Whitechain Sepolia. Prices settle in Inferit Test Credit (ITC), a test token with no monetary value. The contracts are not audited.

Inferit is a peer-to-peer market for AI inference. Anyone can sell inference at their own price, and buyers and agents pay per request through one OpenAI-compatible API. This repository is that API's model list in one JSON file, checked against the live API on 2026-10-13.

Contents

• What is in it
• A row
• Fields
• Licence classes
• Prices and sellers: ask the API
• Use it
• Sell one of these models
• Check it
• Sources
• Licence

What is in it

File	What
catalog.json	26 models: 23 open-weight models and 3 closed-weight reference rows
licenses.json	The 9 licences those models use: name, link and the class the API applies when a seller lists an offer
schema/catalog.schema.json, schema/licenses.schema.json	JSON Schemas (draft 2020-12) for both files
scripts/validate.mjs	The rules a schema cannot express: unique ids and aliases, licence links between the files
scripts/check-live.mjs	Compares catalog.json with the live GET /v1/models

A row

{
  "id": "qwen/qwen-2.5-14b-instruct",
  "name": "Qwen: Qwen2.5 14B Instruct",
  "owned_by": "qwen",
  "aliases": ["qwen2.5:14b-instruct", "qwen2.5:14b", "qwen2.5-14b-instruct", "Qwen/Qwen2.5-14B-Instruct"],
  "context_length": 32768,
  "license": "apache-2.0",
  "license_class": "permissive",
  "open_weight": true,
  "servable": true
}

Fields

The field names are the API's own, so a row here parses like a row of GET /v1/models. license_class is the one field this file adds.

Field	Type	Meaning
id	string	The model id to send as model in POST /v1/chat/completions: publisher/model, lowercase
name	string	Display name
owned_by	string	The publisher: the part of the id before the /
aliases	string[]	Other names the API accepts for the model, matched case-insensitively: an Ollama tag (qwen2.5:14b-instruct) or a Hugging Face repo id (Qwen/Qwen2.5-14B-Instruct). The seller app uses them to match local model names to catalog ids
context_length	integer	The model's context window, in tokens. A seller may serve a smaller window: each offer's own is in GET /v1/models/{id}/endpoints
license	string	The licence of the model's weights, an id in licenses.json
license_class	string	How the API treats an offer on this model when a seller lists it: permissive, conditional or not_allowed (next section)
open_weight	boolean	true when the weights are published; false exactly when license is proprietary
servable	boolean	false for a reference row: no seller can list the model on Inferit, and a request for it answers 404 model_not_servable

Licence classes

Class	When a seller lists an offer	Licences in this catalog
permissive	It lists with no acknowledgement	Apache License 2.0, MIT License
conditional	The seller accepts the licence's conditions with licenseAck: true	Qwen License Agreement, Llama 3.1, 3.3 and 4 Community License Agreements, Gemma Terms of Use, Modified MIT License (Kimi)
not_allowed	The API refuses it (license_not_listable)	Proprietary (closed weights): the 3 reference rows

The classes are Inferit's reading of each licence text, read on 2026-10-02. They are not legal advice: read the licence licenses.json links. The seller is responsible for having the right to serve what it lists; the licence terms are between the seller and the model's lab.

Prices and sellers: ask the API

This file holds no prices, no seller data and no quality scores. Those change while the market runs, so they come from the API:

You want	Ask
List price, the cheapest healthy offer's price, seller counts and health per model	GET /v1/models: pricing, best_price, live
Every seller's offer of one model: all-in price, context window, health	GET /v1/models/{id}/endpoints, with the id URL-encoded (qwen%2Fqwen-2.5-14b-instruct)
One model's order book	GET /api/markets/{model}
The quality ranking model auto uses	GET /v1/routing and inferit.cc/docs/routing

API prices are integers in µ per 1M tokens: one µ is a micro-dollar, a millionth of a dollar. On the testnet it is a micro-ITC, a test token with no monetary value.

GET /v1/models also lists three rows that are not models: auto, auto:cheap and auto:best. They are model auto's policies, where Inferit picks the model and then the cheapest healthy seller of it, so this file leaves them out. The routing docs describe them.

Use it

The file is plain JSON. Read it from this repository:

const url = 'https://raw.githubusercontent.com/InferitAI/model-catalog/main/catalog.json';
const { models } = await (await fetch(url)).json();

// Resolve any name the API accepts (id or alias, any case) to a catalog row.
const byName = new Map(models.flatMap((m) => [m.id, ...m.aliases].map((k) => [k.toLowerCase(), m])));
byName.get('qwen2.5:14b-instruct').id; // 'qwen/qwen-2.5-14b-instruct'

// Models a seller can list today.
const servable = models.filter((m) => m.servable);

Or ask the live API for the same rows with their prices:

curl -s https://api-production-c74b9.up.railway.app/v1/models \
  | jq '.data[] | select(.auto == null) | {id, context_length, license, servable, best_price}'

Sell one of these models

Sellers may list any model at their own price. Today the API accepts an offer only for a model in this catalog; any other model answers model_not_found.

• From your own computer: inferit-seller connects a local Ollama, LM Studio, vLLM or llama.cpp server to the market with one command, with no open port or domain. Start with the seller quickstart.
• From an https endpoint: inferit.cc/docs/sellers.
• Set prices: the Sell workspace, inferit.app/sell.

To ask for a model, open an issue with its Hugging Face repo id and a link to its licence. A model joins the API's catalog first; this file follows it.

Check it

Node 18 or newer. No install needed:

node scripts/validate.mjs      # unique ids and aliases, licence links, open_weight and servable rules
node scripts/check-live.mjs    # every row against the live GET /v1/models (exit 1 on a difference)
npx --yes ajv-cli@5.0.0 validate --spec=draft2020 -s schema/catalog.schema.json -d catalog.json
npx --yes ajv-cli@5.0.0 validate --spec=draft2020 -s schema/licenses.schema.json -d licenses.json

CI runs the schema checks and validate.mjs on every push to main and on every pull request.

Sources

• The rows: Inferit's catalog, as GET /v1/models served it on 2026-10-13. check-live.mjs matched all 26 rows that day.
• Ids, names and context windows: first read from OpenRouter's public model list on 2026-10-02, a data source and not a relationship. The Qwen 2.5 14B row was added by hand.
• Licences: read from each model's card and licence file on 2026-10-02.

Licence

• Data (catalog.json, licenses.json): Creative Commons Attribution 4.0 (CC BY 4.0). Credit it as "Inferit model catalog (github.com/InferitAI/model-catalog), CC BY 4.0".
• Schemas and scripts (schema/, scripts/, .github/): Apache License 2.0.
• Model names belong to their publishers. Each model's weights carry their own licence, which licenses.json links; this repository grants no rights to any model.
• The Inferit name and logo are not covered by either licence: see InferitAI/branding.

More from Inferit

• inferit-seller: sell inference from your own computer (npm inferit-seller).
• whitechain-x402-facilitator: the first public x402 v2 facilitator for Whitechain, live on Whitechain Sepolia (testnet).
• branding: the Inferit logo files, colours and type.
• inferit.cc · docs · X

Inference, by the µ.