Model ids & variants

Model ids on api14 use the form provider/model, for example openai/gpt-5-mini. Pass the id in the model field of any request.

{ "model": "anthropic/claude-sonnet-4.5" }

Finding model ids

• Browse the models page, every model with its id, prices, context length and capabilities.
• Call GET /models from code; filter by category with ?category=image. See Models.

Some examples:

Ids are case sensitive. An unknown or disabled id returns 404 model_not_found.

Variants

A variant is a suffix after a colon that changes how the request is served. Variants work with the same base model; the base model's availability and allowlist rules apply.

{ "model": "openai/gpt-5-mini:online" }

:nitro and :floor are shortcuts for provider sorting, the same as provider: { sort: "throughput" } and provider: { sort: "price" }. See Provider routing & fallbacks.

Model fallbacks

Send a models array to try other models when the first one fails or is unavailable:

{
  "model": "anthropic/claude-sonnet-4.5",
  "models": ["openai/gpt-5-mini", "google/gemini-2.5-flash"],
  "messages": [{ "role": "user", "content": "Hello" }]
}

Every model in the list must be available and allowed for your key and plan, otherwise the request is rejected before it is sent. The response's model field tells you which one answered, and you pay that model's price.

Allowlists and plans

• A key with a model allowlist can only use the listed ids; other ids return 403 model_not_allowed.
• Every plan, Free included, covers all model categories, video and music/audio among them. A plan limited to certain categories returns 403 plan_required for the rest. See Pricing & billing.

Model changes

The catalog updates automatically. New models appear on their own; removed models return 404 model_not_found. Pin exact ids in production and use models fallbacks for resilience.