API reference — Model catalog

The model catalog is the registry of AI models available in Arbitex. Each entry maps a provider (e.g., anthropic, openai) and a model ID (e.g., claude-3-5-sonnet-20241022) to a set of capabilities and cost metadata. Models must be in the catalog (and have is_active: true) before they can be referenced in routing rules or API requests.

All model catalog endpoints are admin-only. Authentication: Authorization: Bearer arb_live_<admin-key>.

Base path: GET|POST|PUT|DELETE /api/admin/models/catalog

Schemas

ModelConfig object

Field	Type	Description
`id`	UUID	Internal record identifier
`provider`	string	Provider name (e.g., `anthropic`, `openai`, `azure`)
`model_id`	string	Provider-specific model identifier
`display_name`	string	Human-readable name for UI display
`is_active`	boolean	Whether the model is available for routing
`is_default`	boolean	Whether this is the default model for its provider
`capabilities`	object	Model capability flags (see below)
`cost_per_input_token`	float	Cost per input token in USD (nullable)
`cost_per_output_token`	float	Cost per output token in USD (nullable)
`created_at`	datetime	ISO 8601 creation timestamp

Capabilities object:

Field	Type	Description
`streaming`	boolean	Supports streaming responses
`function_calling`	boolean	Supports function/tool calling
`vision`	boolean	Supports image inputs
`structured_output`	boolean	Supports structured JSON output
`embeddings`	boolean	Supports embeddings generation
`max_context_window`	integer	Maximum context window in tokens

ModelDiscoveryResponse object

Returned by /discover — includes catalog membership status:

Field	Type	Description
`provider`	string	Provider name
`model_id`	string	Provider-specific model ID
`display_name`	string	Human-readable name
`max_tokens`	integer	Maximum context window
`supports_streaming`	boolean	Streaming support
`already_in_catalog`	boolean	Whether this model is already in the catalog

Endpoints

List catalog entries

GET /api/admin/models/catalog/

Returns paginated ModelConfig records with optional filtering.

Query parameters:

Parameter	Type	Default	Description
`page`	integer	`1`	Page number (1-based)
`page_size`	integer	`20`	Records per page (1–200)
`search`	string	—	Case-insensitive substring match against `display_name` and `model_id`
`provider`	string	—	Exact provider name filter
`is_active`	boolean	—	Filter by active status

Request:

GET /api/admin/models/catalog/?provider=anthropic&is_active=true
Authorization: Bearer arb_live_your-admin-key

Response (200):

{
  "items": [
    {
      "id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
      "provider": "anthropic",
      "model_id": "claude-3-5-sonnet-20241022",
      "display_name": "Claude 3.5 Sonnet",
      "is_active": true,
      "is_default": false,
      "capabilities": {
        "streaming": true,
        "function_calling": true,
        "vision": true,
        "structured_output": true,
        "embeddings": false,
        "max_context_window": 200000
      },
      "cost_per_input_token": 0.000003,
      "cost_per_output_token": 0.000015,
      "created_at": "2026-03-01T00:00:00Z"
    }
  ],
  "total": 12,
  "page": 1,
  "page_size": 20
}

Create a catalog entry

POST /api/admin/models/catalog/

Creates a new ModelConfig record. Returns 409 Conflict if a record with the same provider + model_id already exists.

Request body:

Field	Required	Description
`provider`	Yes	Provider name
`model_id`	Yes	Provider-specific model ID
`display_name`	Yes	Human-readable name
`is_active`	No	Default: `true`
`capabilities`	No	Capability flags object
`cost_per_input_token`	No	Cost per input token in USD
`cost_per_output_token`	No	Cost per output token in USD

Request:

POST /api/admin/models/catalog/
Authorization: Bearer arb_live_your-admin-key
Content-Type: application/json

{
  "provider": "openai",
  "model_id": "gpt-4o-mini",
  "display_name": "GPT-4o mini",
  "is_active": true,
  "capabilities": {
    "streaming": true,
    "function_calling": true,
    "vision": true,
    "structured_output": true,
    "embeddings": false,
    "max_context_window": 128000
  },
  "cost_per_input_token": 0.00000015,
  "cost_per_output_token": 0.0000006
}

Response (201 Created): Returns the created ModelConfig object.

Error (409 Conflict):

{ "detail": "A catalog entry for provider 'openai' and model_id 'gpt-4o-mini' already exists." }

Get a catalog entry

GET /api/admin/models/catalog/{model_config_id}

Response (200): Returns the matching ModelConfig object.

Error (404 Not Found):

{ "detail": "Model catalog entry not found" }

Update a catalog entry

PUT /api/admin/models/catalog/{model_config_id}

Partial update — only non-null fields are applied.

Updatable fields: display_name, is_active, is_default, capabilities, cost_per_input_token, cost_per_output_token.

Request:

PUT /api/admin/models/catalog/3fa85f64-5717-4562-b3fc-2c963f66afa6
Authorization: Bearer arb_live_your-admin-key
Content-Type: application/json

{
  "is_active": false,
  "cost_per_output_token": 0.000018
}

Response (200): Returns the updated ModelConfig object.

Delete a catalog entry

DELETE /api/admin/models/catalog/{model_config_id}

Permanently removes the catalog entry.

Response: 204 No Content

Error (404 Not Found): Entry not found.

Discover models from providers

GET /api/admin/models/catalog/discover

Queries all registered provider adapters for their available models and returns the results annotated with already_in_catalog. Providers that are unavailable or return an error are skipped (no failure — the response includes results from available providers only).

Request:

GET /api/admin/models/catalog/discover
Authorization: Bearer arb_live_your-admin-key

Response (200):

[
  {
    "provider": "anthropic",
    "model_id": "claude-opus-4-6",
    "display_name": "Claude Opus 4.6",
    "max_tokens": 200000,
    "supports_streaming": true,
    "already_in_catalog": false
  },
  {
    "provider": "anthropic",
    "model_id": "claude-3-5-sonnet-20241022",
    "display_name": "Claude 3.5 Sonnet",
    "max_tokens": 200000,
    "supports_streaming": true,
    "already_in_catalog": true
  }
]

Use this endpoint to identify new provider models that have not yet been imported into the catalog.

Sync discovered models into catalog

POST /api/admin/models/catalog/sync

Queries all registered provider adapters (same as /discover) and creates new catalog entries for any provider+model_id pairs not already in the catalog. Existing entries are left unchanged. The operation is idempotent.

Request:

POST /api/admin/models/catalog/sync
Authorization: Bearer arb_live_your-admin-key

Response (200): Returns the list of newly created ModelConfig records. An empty array means all discovered models were already in the catalog.

[
  {
    "id": "abc12345-...",
    "provider": "anthropic",
    "model_id": "claude-opus-4-6",
    "display_name": "Claude Opus 4.6",
    "is_active": true,
    "is_default": false,
    "capabilities": {
      "streaming": true,
      "function_calling": false,
      "vision": false,
      "structured_output": false,
      "embeddings": false,
      "max_context_window": 200000
    },
    "cost_per_input_token": null,
    "cost_per_output_token": null,
    "created_at": "2026-03-12T14:00:00Z"
  }
]

Bulk enable or disable entries

PATCH /api/admin/models/catalog/bulk

Sets is_active on multiple catalog entries in a single operation. UUIDs that do not match any record are silently ignored.

Request body:

Field	Type	Required	Description
`model_ids`	UUID[]	Yes	List of ModelConfig record UUIDs to update
`is_active`	boolean	Yes	Target active state

Request:

PATCH /api/admin/models/catalog/bulk
Authorization: Bearer arb_live_your-admin-key
Content-Type: application/json

{
  "model_ids": [
    "3fa85f64-5717-4562-b3fc-2c963f66afa6",
    "7c8d9e0f-1234-5678-abcd-ef0123456789"
  ],
  "is_active": false
}

Response (200): Returns the list of updated ModelConfig records.

Common workflows

Onboarding a new provider

Configure the provider adapter with credentials in the Platform environment.
Call GET /discover to see available models.
Call POST /sync to import them all.
Update cost metadata for billing accuracy: PUT /catalog/{id} with cost_per_input_token and cost_per_output_token.
Verify models appear as routing targets in the admin console.

Deprecating a model

When a provider retires a model:

PUT /catalog/{id} with { "is_active": false } — existing routing rules will fail over to alternatives.
Update routing rules that reference the deprecated model.
Optionally DELETE /catalog/{id} to remove the record.

API reference — Model catalog

Schemas

ModelConfig object

ModelDiscoveryResponse object

Endpoints

List catalog entries

Create a catalog entry

Get a catalog entry

Update a catalog entry

Delete a catalog entry

Discover models from providers

Sync discovered models into catalog

Bulk enable or disable entries

Common workflows

Onboarding a new provider

Deprecating a model

See also