NAnace
Evaluate

Evaluate questions (OpenAI Decisions format)

POST /v1/decisions takes OpenAI's Decisions wire format on the same models as POST /v1/systemone, so the OpenAI SDKs, LiteLLM and Vercel AI Gateway call Drex with only a base URL and model change. The same key, funding, rate limits, validation and billing apply; only the request and response shapes differ. See OpenAI-compatible decisions.

  • Billing: only a 200 is billed, at the per-input-token price, exactly as /v1/systemone bills the same request.
  • Accepted and ignored: safety_identifier, prompt_cache_key, prompt_cache_options, prompt_cache_breakpoint and an image's detail are checked against OpenAI's schema and then dropped. Any other key Drex doesn't know is ignored.
  • Errors: Drex's error.type and status, plus OpenAI's error.code and error.param.

Checks run in the same order as /v1/systemone and stop at the first failure.

post/v1/decisions

Request

Request body

modelstring

As on POST /v1/systemone. A registered id or alias is served as named, and no model is served like drex-latest. Any other name (gpt-6-luna, jev-latest, a mistyped drex- name) or a registered id that isn't available yet returns 422 with code: model_not_found and param: model.

safety_identifierstring nullable

OpenAI's opaque end-user identifier. Accepted for compatibility, up to 128 characters; Drex neither forwards nor stores it.

prompt_cache_keystring nullable

Accepted for OpenAI compatibility, up to 64 characters, and ignored: Drex has no prompt cache.

Example request

{
  "model": "drex-max",
  "input": [
    {
      "content": [
        {
          "image_url": "data:image/png;base64,iVBORw0KGgo..."
        }
      ]
    }
  ]
}

Response

Every question answered, in order. usage.input_tokens is debited from the account's balance before the response is returned.

model'drex-v1.5' | 'drex-v2.0' | 'drex-max' | 'drex-max-v1.1' | 'drex-max-finance' required

The version that answered.

request_idstring required

Identifier of this request, req_ plus 32 hex characters. Also sent in the x-request-id header, and on /v1/systemone and /v1/models in x-typesafe-request-id too. Quote it in support requests.

Example response

{
  "request_id": "req_bd2dffad68f5f5fa4ce702e9bd516fb4"
}

Changes

No recorded changes to this operation across all 1 revision of this API.