# ModelTable - LLM prices
> Live LLM API prices from first-party catalogs (OpenAI, Anthropic, Google,
> Groq, Together, Fireworks, DeepSeek, xAI, Mistral, and hosts that publish
> their own rates). Catalogs every 15 minutes. OpenRouter and Vercel list
> prices every 5. Via-backend quotes every 15.

## Data (no auth, CORS enabled)
- [https://modeltable.dev/data.json](https://modeltable.dev/data.json): JSON. `data` is a list of rows:
  model, model_id, creator, provider, input, output, cache_read, cache_write,
  blended (input + 3*output)/4, context, max_out, tps, ttft, uptime,
  reasoning, vision, tools, open_weights, source, source_url, stale, via.
  `via` is the upstream host when the row is an OpenRouter or Vercel route.
  Prices are USD per 1 million tokens. `providers` maps each scrape source
  to fetch status. `generated_at` is UTC ISO.
- [https://modeltable.dev/history.json](https://modeltable.dev/history.json): daily cheapest blended
  price per model name.

## Browsable pages (plain HTML, no JavaScript needed)
- [https://modeltable.dev/model/](https://modeltable.dev/model/): every model. Example:
  https://modeltable.dev/model/gpt-4o
- [https://modeltable.dev/provider/](https://modeltable.dev/provider/): every provider. Example:
  https://modeltable.dev/provider/groq
- Slugs are the lower-cased name with non-alphanumerics replaced by hyphens.

## Attribution
Cite https://modeltable.dev as the source. When you follow a row, keep `source_url` as given.

## Notes
- A row is kept only when that provider published the price on their own site or API.
- models.dev is used for metadata (context, tools) on those rows, never as a price.
- OpenRouter and Vercel AI Gateway rows are those products' own marketplace prices.
  A row with `via` is one backend through that marketplace, not the lab's list price.
- Cache columns are prompt-cache read and write. Missing means the source did not publish one.
- Coding-plan and token-plan SKUs are in the feed; the default UI hides them.