Endpoint Reference: Arena All Models

These endpoints return Cranium's All Models catalog: every model Cranium tracks, not only the models discovered in your tenant. The catalog is the same for every tenant, except aiSystemCount, which counts your AI Systems that use the model. To list only the models discovered and tested in your tenant, use the Arena Models endpoint instead.

List All Models

Returns a paginated list of the All Models catalog. Use it to load the full catalog into your own systems and join it to your other Arena data by modelId.

Request

GET /api/public/arena/allmodels

Authentication

Bearer token. See Authentication & Generating Credentials.

Required Permission and License

Api_ArenaAllModels_Read. These endpoints are available only to tenants licensed for All Models. A missing license and a missing permission return the same 401 UNAUTHORIZED response.

Have an administrator grant Api_ArenaAllModels_Read to your API client before your first call.

Query Parameters

Parameter	Type	Value
`limit`	integer	Results per page. Default 20. Maximum 100. A value of 0 or less is treated as 1.
`cursor`	string	Sync cursor. Omit on the first call to begin from the beginning. Pass `nextCursor` from the previous response on subsequent calls.
`updatedAfter`	ISO 8601 timestamp	Return only models added at or after this timestamp. Must be between 1753-01-01 and 9999-12-31. Cannot be sent with `cursor`, even when `cursor` is empty.

Response Fields

Field	Type	Rules
`modelId`	GUID	Unique identifier for the model. The same identifier as `modelId` on the Arena Models and Arena Model Vulnerabilities endpoints, and as `arenaModelId` on the Bill of Materials models feed. Use it to join the catalog to those endpoints.
`modelName`	string	Model name.
`modelFamilyId`	GUID	Identifier for the model family this model belongs to.
`modelFamilyName`	string	Family name, for grouping related models.
`modelSource`	string	Where the model comes from. The value is free text from the catalog. Today it is `hf` (Hugging Face) for catalog models and an empty string for models added through a pen-test upload. Treat as an open string.
`modelType`	string	Model type. One of `GenAI` (generative AI), `DL` (deep learning), `TradML` (traditional machine learning) or `Unknown`.
`referenceUrl`	string	Link to the model's reference page.
`discoveredAt`	ISO 8601 timestamp	When the model was first discovered. UTC without a trailing `Z`.
`updatedAt`	ISO 8601 timestamp	When the model record was last modified. UTC without a trailing `Z`. This is the field the sync cursor tracks.
`modelStatus`	string	Testing state. One of `Untested`, `Queued`, `InTesting`, `Completed`, `Failed` or `NotTestable`.
`weaknessCount`	integer	Number of identified weaknesses.
`attackCategoriesVulnerableCount`	integer	Number of attack categories with at least one successful attack.
`knownRisksCount`	integer	Number of known risks associated with the model.
`automatedTestCount`	integer	Number of automated penetration tests run.
`manualTestCount`	integer	Number of manual penetration tests run.
`aiSystemCount`	integer	Number of your AI Systems that use the model. This is the only value in the catalog that differs by tenant.
`vulnerabilityLikelihoodPercentage`	integer or null	The model's overall risk score as a percentage: the average attack success rate across all tested attack categories, with each category counted equally. Null when the model has not been tested.
`severity`	string	Severity, set from `vulnerabilityLikelihoodPercentage`. One of `Low` (under 10%), `Medium` (10% to 50%), `High` (above 50% up to 95%) or `Critical` (above 95%).

Keeping the Catalog Current

updatedAfter returns the models added since that time. Changes to a model's status, severity and counts do not move it in the feed, so an incremental sync does not pick them up. Re-read the full catalog periodically to refresh those values.

Sample Response

{
  "data": [
    {
      "modelId": "7e1d9c3b-2a4f-4b8e-9d6c-5f0a1b2c3d4e",
      "modelName": "gpt-4o",
      "modelFamilyId": "1a2b3c4d-5e6f-4a7b-8c9d-0e1f2a3b4c5e",
      "modelFamilyName": "GPT-4o",
      "modelSource": "hf",
      "modelType": "GenAI",
      "referenceUrl": "https://platform.openai.com/docs/models/gpt-4o",
      "discoveredAt": "2026-03-04T09:12:44.218734",
      "updatedAt": "2026-10-18T22:05:31.904512",
      "modelStatus": "Completed",
      "weaknessCount": 6,
      "attackCategoriesVulnerableCount": 4,
      "knownRisksCount": 2,
      "automatedTestCount": 120,
      "manualTestCount": 8,
      "aiSystemCount": 3,
      "vulnerabilityLikelihoodPercentage": 18,
      "severity": "Medium"
    },
    {
      "modelId": "4c8a1e2d-7b3f-4d9a-b6e1-2f0c5a8d3b97",
      "modelName": "llama-3.1-8b-instruct",
      "modelFamilyId": "9f8e7d6c-5b4a-4392-8170-6f5e4d3c2b1a",
      "modelFamilyName": "Llama 3.1",
      "modelSource": "hf",
      "modelType": "GenAI",
      "referenceUrl": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct",
      "discoveredAt": "2026-05-21T16:40:02.551093",
      "updatedAt": "2026-10-11T08:17:56.330871",
      "modelStatus": "Untested",
      "weaknessCount": 0,
      "attackCategoriesVulnerableCount": 0,
      "knownRisksCount": 1,
      "automatedTestCount": 0,
      "manualTestCount": 0,
      "aiSystemCount": 0,
      "vulnerabilityLikelihoodPercentage": null,
      "severity": "Low"
    }
  ],
  "pagination": {
    "limit": 2,
    "nextCursor": "eyJ1cGRhdGVkQXQiOiIyMDI2LTEwLTExVDA4OjE3OjU2LjMzMDg3MTAiLCJpZCI6IjRjOGExZTJkLTdiM2YtNGQ5YS1iNmUxLTJmMGM1YThkM2I5NyJ9",
    "hasMore": true
  },
  "error": null,
  "meta": {
    "requestId": "a9c2e5f1-3d7b-4e0a-8c6f-1b4d9e2a7c35",
    "timestamp": "2026-10-22T15:04:14.3890000Z"
  }
}

modelId is the join key to the Arena Models and Arena Model Vulnerabilities endpoints. gpt-4o has been tested and is used by three of your AI Systems. llama-3.1-8b-instruct has not been tested, so vulnerabilityLikelihoodPercentage is null, and none of your AI Systems use it. hasMore is true, so pass nextCursor as cursor on the next call.

Error Responses

400 BAD_REQUEST: the request was rejected. The message names the reason. A 400 is returned when:
- limit is over 100: limit must be 100 or less.
- updatedAfter is outside the allowed range: updatedAfter must be between 1753-01-01 and 9999-12-31.
- cursor and updatedAfter are both sent, even when cursor is empty: cursor and updatedAfter are mutually exclusive.
- The cursor is not valid.
- The request includes a page parameter.
401 UNAUTHORIZED: the request is missing a token, the token lacks Api_ArenaAllModels_Read, or the tenant is not licensed for All Models.
500 INTERNAL_ERROR: an unexpected server error.

Every error response includes error.traceId. Include it when you report an issue to Cranium support.

Sample 401 response:

{
  "data": null,
  "pagination": null,
  "error": {
    "code": "UNAUTHORIZED",
    "message": "Unauthorized.",
    "traceId": "0HNOVC2K1R7QF:00000001"
  },
  "meta": {
    "requestId": "e1a5c9d3-4f7b-4a2e-8c6d-0b3f9e1a5d72",
    "timestamp": "2026-10-22T15:04:15.9830000Z"
  }
}

Get an All Models Entry

Returns one entry from the All Models catalog. Use it to look up a single model by the modelId you received from another endpoint.

Request

GET /api/public/arena/allmodels/{modelId}