Skip to main content
A model is served by a provider inside a workspace. Reads are scoped by workspace_id. Reads need models:read and writes need models:write.
Creates always land in your account’s default workspace. A workspace_id in the body is ignored.

List models

string
Limit results to one workspace.
string
Limit results to one provider.
string
One of chat, embedding, moderation, vision, audio, or image.
string
Match models whose name contains this value.
Send gateway_name to the gateway to route inference to this model.
display_name, gateway_name, input_cost, output_cost, and context_window are all nullable on a stored model, even though they are required when you create one. Models imported from the catalog can arrive with gaps.

Cost fields

Beyond input_cost and output_cost, a model may carry cache_read_cost, cache_write_cost, input_cost_over, output_cost_over, cost_over_threshold, request_cost, per_image_cost, per_minute_cost, vision_input_cost, audio_input_cost_per_min, and audio_output_cost_per_min. Each is nullable.

Capability flags

supports_vision, supports_audio_input, supports_audio_output, supports_thinking, supports_redacted_thinking, supports_tool_streaming, supports_computer_use, supports_citations, attachment, reasoning, and tool_call are booleans. reasoning_style, speed_mode, audio_formats, image_sizes, and max_images_per_request describe how those capabilities behave.

Get a model

Create a model

string
required
The provider that serves the model.
string
required
4 to 32 characters. Letters, numbers, hyphens, underscores, and dots. No slashes or spaces.
string
required
A readable label, 4 to 64 characters.
string
required
One of chat, embedding, moderation, vision, audio, or image.
number
required
Cost per input unit.
number
required
Cost per output unit.
integer
required
The context window size in tokens.
string
The name the upstream provider expects, if it differs from name.
number
Required when type is image.
number
Required when type is audio.
number
Cost per cached input unit.
number
Cost to write to the cache.
integer
The largest output the model will produce.
boolean
Whether the model accepts images.
boolean
Whether the model accepts audio.
boolean
Whether the model returns audio.
string
The audio formats the model accepts.
string
The image sizes the model produces.
integer
A cap on images per request.

Update a model

PATCH is a full replace, not a merge. It validates the same schema as create, so provider_id, name, display_name, type, input_cost, output_cost, and context_window are all required on every update.
The response carries no body beyond { "success": true }. The SDK re-reads the model after a write.

Delete a model