Skip to main content
Model ids have the form provider/model. Use them in dari-gateway.yaml and in requests as dari/model/<id>. The list below shows the current generation of each model family. The gateway’s bundled catalog moves with each gateway release, so when a newer model ships, update your config to use it.

Current Models

Claude Fable is the top Claude tier, above Opus. Meta models are served through a Muse Code subscription, which serves only Muse Code clients, or through META_API_KEY. Gemini models are served through GEMINI_API_KEY. Anthropic models served through a Claude Code subscription serve only Claude Code clients. See Subscriptions. A conversation that contains a PDF routes only to models that accept PDFs, and fails with a 400 if the router has none. Text documents and search results from clients such as Claude Code reach every model as text. Thinking levels differ by model. Use medium or high when you want a value every model on this list accepts, and check the rest with the catalog below.

List The Live Catalog

The table above is a snapshot. For the live list, with each model’s context window, output limit, accepted thinking levels, and list price, call the models route with any gateway key:
Each entry includes:
  • id: the id to use as provider/model and in dari/model/<id>.
  • context_window and max_output_tokens.
  • thinking_levels: the levels the model accepts.
  • input_modalities.
  • price: list price in USD per million tokens, with cached-input and cache-write rates and long-context tiers. Prices change, so use the endpoint rather than copying them.
  • price_as_of: when the price was read.
Narrow the list to some providers with provider=:
The default providers are OpenAI, Anthropic, Meta, xAI, Google, and Fireworks, read from the bundled catalog. OpenRouter and Vercel AI Gateway are read live from their public model endpoints, so their entries change without a gateway release.

Other Providers

For a model that is not in the catalog, such as a model on your own server, add a provider endpoint and use its name as the prefix. See Custom Provider Endpoints.