modelfax

A machine-readable registry of AI model pricing, context windows, and lifecycle dates (release / deprecation / retirement) across OpenAI, Anthropic, and Google, published as a static site with a free JSON API.

Upstream provider pages are re-read daily; every change lands as a reviewed pull request, so git log -- data/ is the price history.

• Site: https://bytebrujo.github.io/modelfax/
• Cost calculator: https://bytebrujo.github.io/modelfax/calculator.html

Try it

# every Anthropic model, with prices per million tokens
curl -s https://bytebrujo.github.io/modelfax/data/anthropic.json | jq '.models[] | {model_id, input: .pricing.input_per_mtok, output: .pricing.output_per_mtok}'

# everything retiring, across all three providers
curl -s https://bytebrujo.github.io/modelfax/data/{openai,anthropic,google}.json \
  | jq -s '[.[].models[]] | map(select(.dates.retired)) | sort_by(.dates.retired) | .[] | {id, retired: .dates.retired}'

JSON API

The API is the data files, served as static files. CORS is open.

URL	Contents
https://bytebrujo.github.io/modelfax/data/openai.json	OpenAI models
https://bytebrujo.github.io/modelfax/data/anthropic.json	Anthropic models
https://bytebrujo.github.io/modelfax/data/google.json	Google (Gemini API) models
https://bytebrujo.github.io/modelfax/schema/model.schema.json	JSON Schema 2020-12 for every data file

Each data file is {"schema_version": "1.0.0", "models": [ … ]}. One record:

{
  "id": "anthropic:claude-sonnet-5",
  "provider": "anthropic",
  "model_id": "claude-sonnet-5",
  "display_name": "Claude Sonnet 5",
  "family": "claude",
  "status": "available",
  "modalities": { "input": ["text", "image"], "output": ["text"] },
  "context_window": 1000000,
  "max_output_tokens": 128000,
  "pricing": {
    "currency": "USD",
    "input_per_mtok": 2.0,
    "output_per_mtok": 10.0,
    "cached_input_per_mtok": 0.2,
    "batch_input_per_mtok": 1.0,
    "batch_output_per_mtok": 5.0
  },
  "dates": { "released": null, "deprecated": null, "retired": null },
  "sources": ["https://platform.claude.com/docs/en/about-claude/pricing"],
  "last_verified": "2026-09-04",
  "notes": "…"
}

Field reference:

field	type	notes
id	string	<provider>:<model_id>, globally unique
provider	enum	openai · anthropic · google
model_id	string	the exact string the provider's API accepts
status	enum	announced · available · deprecated · retired
modalities	object	input/output arrays from text, image, audio, video
context_window	integer	tokens
max_output_tokens	integer or null	null when the provider does not publish it
pricing	object or null	USD per million tokens; null only for announced/retired
dates	object	released, deprecated, retired: ISO date or null
sources	array	https URLs on the provider's own domain where the data was read
last_verified	date	when the record was last confirmed against sources
notes	string	free text, ≤500 chars (tiered pricing, aliases, caveats)

Stability promise

• Paths above never change within a schema major version.
• Additive schema changes bump the minor version (1.1.0); consumers should ignore unknown fields. Breaking changes bump the major, ship a migration script, and are announced in AGENTS.md.
• schema_version in each data file always equals the const in the schema.

Coverage

Each provider module declares the model ids the registry covers. Upstream pages list many more, including image, video and embedding models that have no per-token price and do not fit this schema. A model that appears on a provider's pricing or models page without being covered is reported by the scraper as an untracked model rather than being recorded with missing fields, and widening coverage is a deliberate change. Open an issue or a PR if a model you need is missing.

Context windows, max output tokens and modalities are only published in a scrapeable table by one of the three providers, so those three fields are maintained by hand and everything else is scraped.

Update cadence

A scheduled job scrapes every provider daily at 06:00 UTC. Changes are opened as a pull request and merged after CI passes. When a provider page changes shape, the scraper fails loudly and stale-but-valid data stays published until a fix lands; last_verified tells you how old each record is.

Local development

Requires Node 24 (see .nvmrc) and make.

npm ci
make check      # lint · schema · test · scrape-dry (fixtures only) · site-smoke
make scrape     # live scrape; exit 3 = data changed, 4 = parse failure, 5 = fetch failure
make serve      # assemble the Pages layout into _site/ and serve on :8000

make check needs no network beyond npm ci. Read AGENTS.md before changing anything.

Repository layout

schema/model.schema.json   the contract
data/<provider>.json       the API
scrapers/run.js            orchestrator; scrapers/providers/*.js one module per provider
fixtures/<provider>/       saved upstream pages the parser tests run against
test/                      schema, cross-file invariants, parser, and site smoke tests
site/                      hand-written HTML/CSS/JS, no build step
.github/workflows/         ci (PR gate), scrape (daily), pages (deploy)

Contributing

Corrections are welcome as pull requests against data/ (cite the provider URL in sources) or against a parser (include the new fixture and test in the same PR). CI must be green.

Licenses

Code is licensed under the MIT License. Data in data/ is licensed under CC-BY-4.0; attribute "modelfax" with a link back.

Prices are transcribed from provider documentation and may lag or contain errors. Confirm against the linked sources before relying on them.