AI Models Catalog

A structured, machine-readable catalog of AI model providers and their models. All data is sourced from first-party APIs and official documentation — no third-party aggregators.

Data Format

Model data is stored as YAML files under providers/<provider>/models/. Each file represents one model with its snapshots:

id: gpt-4o
name: GPT-4o
family: gpt-4o
reasoning: true
tool_call: true
attachment: true
structured_output: true
pricing:
  input: 2.5
  output: 10
  cache_read: 1.25
limit:
  context: 128000
  output: 16384
modalities:
  input:
    - text
    - image
  output:
    - text
knowledge: "2023-10"
release_date: "2024-05-13"
last_updated: "2024-08-06"
snapshots:
  - id: gpt-4o-2024-08-06
  - id: gpt-4o-2024-05-13
    deprecated: true

Pricing Types

Type	When	Example
TokenPricing	Per-million-token pricing	input: 2.5, output: 10
VideoPricing	Per-second pricing	unit: per_second, price: 0.03
UnitPricing	Per-image or per-request	unit: per_image, price: 0.04
FreePricing	No cost	unit: free

See types/pricing.ts for the full type definitions.

Usage

Install Dependencies

npm install

Sync Model Data

Fetch the latest model data from a provider's first-party source:

# Sync a specific provider
npx tsx scripts/sync.ts openai
npx tsx scripts/sync.ts anthropic

# Sync all providers
npx tsx scripts/sync.ts

Validate Model Data

Validate all YAML files against the Zod schemas:

npx tsx scripts/validate.ts

Project Structure

├── providers/
│   ├── openai/
│   │   ├── scrape.ts          # Data acquisition from OpenAI's website
│   │   └── models/            # YAML model data files
│   └── anthropic/
│       ├── scrape.ts          # Data acquisition from Anthropic's website
│       └── models/            # YAML model data files
├── types/
│   ├── model.ts               # Model and Snapshot type definitions
│   ├── pricing.ts             # Pricing type definitions
│   ├── provider.ts            # Provider type definitions
│   ├── schemas.ts             # Zod runtime validation schemas
│   └── index.ts               # Re-exports
├── scripts/
│   ├── sync.ts                # Orchestration: scrape → write YAML
│   ├── validate.ts            # Validate all YAML against schemas
│   └── lib/                   # Shared utilities (defineModel, defineProvider, writer)
└── docs/
    ├── data-acquisition.md    # How we acquire and update model data
    └── lessons-learned.md     # Design principles and pitfalls

Adding a New Provider

1. Create providers/<id>/scrape.ts with a scrape() function that returns { provider, models }
2. Data must come from a first-party source (provider's API or website)
3. Include a discovery step — no hardcoded model ID lists
4. Run npx tsx scripts/sync.ts <id> to generate initial data
5. Validate with npx tsx scripts/validate.ts

See docs/data-acquisition.md for detailed guidelines.

Design Principles

• First-party data only — all model data comes from the provider's own API or website
• Dynamic discovery — scrape functions discover models from the source, not from hardcoded lists
• Include deprecated, exclude retired — deprecated models are included with a deprecated: true flag; retired (inaccessible) models are excluded
• Never fabricate data — if required data is missing, skip the model with a warning rather than filling in guessed values
• YAML source format — human-readable, supports comments, machine-parseable
• Snapshot inheritance — dated model versions are nested within the parent model, inheriting all fields

See docs/lessons-learned.md for the full set of design principles and pitfalls.

License

MIT