Jev: Get Structured Evaluation Results from State

Jev receives a state data object and named questions, and returns answers of type boolean, choice or score. When your application already has JSON state to evaluate and you want to turn specified criteria into answers your program can read, you can use Jev.

What Jev Is Suited For

Use Jev when you hand support records, review results or other JSON state to named questions for evaluation, and you need fixed answer types. Each question specifies type and instructions; choice questions may include options and descriptions, and score questions may include criteria listed in order.

For multi-turn interaction, free-form explanation, content writing or rewriting, use a chat model. Jev's responses center on the structured answers corresponding to the named questions.

A cURL Example You Can Copy and Run

First set BAZAARLINK_API_KEY in your shell. The request below submits state data and three question types directly:

curl --include --fail-with-body "${BAZAARLINK_BASE_URL:-https://api.bazaarlink.ai}/v1/evaluations" \
  -H "Authorization: Bearer ${BAZAARLINK_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "typesafe-ai/jev",
  "state": "The support agent issued a full refund and apologised for the delay.",
  "questions": {
    "refunded": {
      "type": "boolean",
      "instructions": "Was a refund issued?"
    },
    "tone": {
      "type": "choice",
      "instructions": "Tone of the agent?",
      "criteria": {
        "apologetic": "The agent apologises",
        "neutral": "No emotional language",
        "hostile": "The agent is rude"
      }
    },
    "quality": {
      "type": "score",
      "instructions": "How complete was the resolution?",
      "criteria": [
        "poor",
        "partial",
        "complete"
      ]
    }
  }
}'

BAZAARLINK_BASE_URL defaults to https://api.bazaarlink.ai. This endpoint uses Jev's evaluations request format. The current API Reference provides a cURL example and does not provide an SDK method that can call this endpoint directly.

Response Format

A successful response includes id, object, model, answers, usage, rounding and warnings. answers uses the question names from the request as keys. A boolean answer includes probability; a choice answer includes the selected choice, along with probabilities keyed by option name. The score of a score answer is an expected value calculated in the order of the criteria array, between index 0 and N−1; its probabilities are keyed by string indexes (for example "0", "1"), corresponding to the criteria order. The score in this example is 1.98. The top-level rounding describes the number of decimal places for probabilities and scores; warnings lists any warnings for this response as an array, which is empty in this example. Usage is expressed by input_tokens, output_tokens and total_tokens.

The following is an example of the actual response format as of 2026-09-25:

{
  "id": "eval-a3d350e5-…",
  "object": "evaluation",
  "model": "typesafe-ai/jev",
  "answers": {
    "refunded": {
      "type": "boolean",
      "probability": 0.99
    },
    "tone": {
      "type": "choice",
      "choice": "apologetic",
      "probabilities": {
        "apologetic": 1,
        "neutral": 0,
        "hostile": 0
      }
    },
    "quality": {
      "type": "score",
      "score": 1.98,
      "probabilities": {
        "0": 0,
        "1": 0.02,
        "2": 0.98
      }
    }
  },
  "usage": {
    "input_tokens": 394,
    "output_tokens": 74,
    "total_tokens": 468
  },
  "rounding": {
    "probabilityDecimals": 2,
    "scoreDecimals": 2
  },
  "warnings": []
}

Pricing and Limits

As of 2026-09-25, the catalog price is USD $0.042 per 1M input tokens and USD $0 per 1M output tokens. usage still reports input and output token counts separately.

The context limit is 32K tokens, counted over the serialized state and questions.

API Reference

For the complete request fields, response examples and error descriptions, see the Jev evaluation API reference.

Related: Jev API pricing and how to call it

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API