Skip to content

Typesafe_Jev_1.14

Model Overview

Item value
Model ID (model) jev-1.14.0
Release Date 2026-10-14
context window length By default, a single request supports 64k tokens and a single query supports 32k tokens. When the request header TypeSafe-Danger-Zone-Long-Context: trueis configured, both values will be increased to 262k tokens. Long context comes with slower processing speed and lower accuracy, while the token-based unit price remains the same as that of the default context.
Maximum Input It is constrained by both per_request_context (stateplus all questions) and per_question_context (stateplus any single question). There is no separate upper limit on the number of questions
Maximum Output The source document does not specify an output token cap. The response returns usage.output_tokens, and output tokens are not charged.
Input Modality Text: stateis of type string, JSON object, or array. Audio and video are not supported. Image support is an experimental capability (documented as Dreamer-ready, with format and accuracy subject to change across versions); both Jev 1.13 and Jev 1.14 can receive images when TypeSafe-Danger-Zone-Images: trueis configured; Box requires Jev 1.14
output modality Structured Answer: noul, choice, score, box. No natural language generated.
Enter Price $0.042 per million input tokens ($42 per billion input tokens). Charges apply only to input tokens
language Primarily supports English with the highest accuracy; other languages (including Chinese, Japanese and Korean) are processable but with lower accuracy, which requires verification with your own data and attention to confidence
Rate Limiting Configured per organization, the dimensions are tokens/s and requests/s, with specific values available in the Developer Console. A 429 will be returned upon exceeding the limit, and the response may include retry-after

Request Protocol

https

Parameter Name Type Required Description
Content-Type string Yes application/json
Authorization string Yes Bearer YOUR_API_KEY

TypeSafe System One

Request URL

POST https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/systemone

Header (Additional to This Agreement)

Parameter Name Type Required Description
TypeSafe-Danger-Zone-Images string No The only allowed value true. Required by the vendor when the request contains images or type= box, otherwise a 400 error will be returned. Disabled by default.
TypeSafe-Danger-Zone-Long-Context string No
The only allowed value true. When enabled, both the per-request and per-question context size is 262k tokens. Disabled by default. Supported in both Jev 1.14 and Jev 1.13.

Request Body Parameters

Parameter Name Secondary Parameters Three-level parameter Band-4 Parameters Type Required Description
model - - - string Yes Fixed to MaaS_Typesafe_Jev_1.14 (vendor's original model: jev-1.14.0). The vendor also accepts the aliases jev-latestand jev-preview, both of which currently point to jev-1.14.0
state - - - string | object | array Yes Content evaluated jointly by all questions will only be assessed once. The string type is suitable for single-segment text; the object type is suitable for records with field names; the array type is suitable for sequences of messages or records. You can use backticks to reference field paths, for example `ticket.subject`. It is independent of the questions: the content goes in state, and the judgment logic goes in questions. Audio and video are not supported. The maximum nesting depth is 64 layers.
questions - - - object Yes At least one key. The key is the question ID defined by the caller, which is returned as-is as the key of answersand not sent to the model. Multiple typecan be mixed in the same request, and each question is evaluated independently. All questions plus the statemust fall within the context of a single request; any single question plus the statemust fall within the context of a single question
{key} - - object Yes One question. typedetermines the remaining fields
type - string Yes noul: Yes/No probability. choice: Select one from unordered options. score: Rate on an ordered scale. box: Locate a target in the image (requires Jev 1.14 and Images Header; no "not found" result, returns four edges even if the target is not in the image)
instructions -
string | object | array | null No The question or statement to be judged. The schema can be omitted or set to null, and it is filled in all document examples. object/array is used to separate the question from the reference data, with field names enclosed in backticks. The maximum nesting depth is 64 layers. boxNo interrogative sentence is required; just state the target and the image, for example, the coffee mug in \photo\``. The image reference is written in the string and only belongs to this question.
criteria - object | array View type noul: Optional, specifies what yes/no means. choice: Required, 1 to 255 options. score: Required, an array of ordered levels; the API supports a minimum of 1 item and a maximum of 10 items, and the documentation recommends at least 2 levels. box: This field is not present. Maximum nesting depth is 64 layers. Object keys (option names, question IDs) will not be replaced by image references.
true string | object | array | null No Only noul. What is yes. Embedded image references are supported
false string | object | array | null No Only noul. What exactly is "no". Supports embedded image references
{optionName} string | object | array | null Yes Only choice. The key is the option name (the response choicereturns this key), and the value indicates when this option applies. When the value is null, it is interpreted solely by the option name. Both the option name and its description will be sent to the model. If the suggested list in the documentation may not be exhaustive, add otheror none of the above. The names of the structure fields are defined by the caller and are not reserved words.
assets - - - object No Image resources, with keys as reference strings. When this string appears in the string values of state, instructions, and criteria, it will be replaced with an image (object keys are not replaced). Each reference must appear at least once, otherwise a 400 error will be returned. Unreferenced, empty references, or references consisting solely of whitespace will result in a 400 error. A single reference can be up to 1024 bytes in UTF-8 encoding, and all references combined can be up to 65536 bytes in total. Short references may accidentally match common words, so the documentation recommends using a new UUID for each image.Each occurrence of the same reference will be counted as an additional image on the model side, and the 32 MiB after expansion will be included in the count. There must be at least 1 key. A vision model and an Images Header are required.
{key} - - object Yes An image. partand data_base64are mutually exclusive; if both are present, the manufacturer shall use part
type - string No If filled, it can only be image
format
- string depending on the upload method png, jpg, jpeg (jpegis equivalent to jpg), in lowercase. Required when used with data_base64; optional when used with part. The actual format shall be detected based on bytes
data_base64
- string Depending on the upload method Standard Base64 (with =padding), no line breaks, no whitespace, no data:prefix. Required when partis not set
part
- string Depending on the upload method The name of the image part in multipart. If set, data_base64will be ignored. Each file part must be referenced by only one asset.

score criteriais an array rather than a named object. Its elements are of type string, object or array, with subscripts starting from 0, and the subscript corresponds to the level. The response scoreis the weighted average of each level number weighted by probability, which may fall between two integer levels. The model only refers to the level description and cannot see the level number; the description should clearly specify the scenario, rather than simply stating "0/1/2" or "more severe than the previous level".

Image restrictions (for vendors, covering both ends): static PNG or JPEG images only, animated PNGs are rejected; maximum size per single file is 16 MiB (for Base64, calculated by post-decoding size); width and height each range from 280 to 1568 pixels; total pixel count ranges from 78400 to 2458624; images will not be auto-scaled. The request body, all images after processing, and references after expansion each have a maximum size of 32 MiB. If any single image is invalid, the entire request will fail. EXIF orientation will not be applied, so pixels must be pre-adjusted to the correct orientation.Image tokens are counted towards usage.input_tokensand the context, and there is no separate unit price. The source document states that the token estimation method has not yet been provided.

multipart rule (recommended by the vendor to avoid Base64 amplification): the first part is named request, with Content-Type: application/json; the Content-Type of each image part can be image/png, image/jpeg, image/jpg, application/octet-streamor omitted, and a 400 error will be returned for other types. requestis a reserved name.The current MaaS entry is not bound according to multipart.

Response Body Parameters

Parameter Name Secondary Parameters Three-level parameter Band-4 Parameters Type Required Description
model - - - string Yes Returned via the MaaS gateway, this field indicates the MaaS model name in the request MaaS_Typesafe_Jev_1.14. When the vendor is directly connected, this is the actual response version; alias requests will be resolved to jev-1.14.0, which may differ from the alias in the request
answers - - - object Yes At least 1 key, the key matches the key of the request questions. typeis the same as that of the corresponding question
{key} - - object Yes The answer to a question
type - string Yes noul、choice、score 或 box
noul - number View type Only noul. The probability of yes, ranging from 0 to 1. A value close to 1 means yes, close to 0 means no, and close to 0.5 means uncertain. There is no separate confidence. It is not a degree scale.
choice - string View type Only choice. probabilitiesthe option name with the highest value
score - number View type Only score. The expected value of the level number, ranging from 0 to the highest level number, can be a decimal. The same score can be obtained from different probability distributions, so it is necessary to check both probabilitiesand confidence
confidence - number | null View type choice, scorerequired, ranging from 0 to 1, calculated based on the concentration degree of the probability distribution, and does not guarantee correctness. boxrequired but can be null (null when the model does not provide confidence), ranging from 0 to 1, with 1 indicating higher certainty. No recommended Box threshold is provided in the source document
probabilities - object View type choice: Probabilities of each option, with keys being the option names. score: Probabilities of each level, with keys being level number strings, consistent with legend. Values range from 0 to 1, and the sum is approximately 1. nouland boxdo not have this field.
legend - object View type Only score. A mapping of level number strings to the request criteriadescription, for interpreting the score. Image references in the description remain as reference strings in the response, not as image bytes
box - object View type Only box. All four edges are the desired positions relative to the image size, ranging from 0 to 1. The origin is at the top left;left< right, top< bottom
top number Yes Top, the proportion of the image height, measured from the top
bottom number Yes Bottom, the proportion of the image height, measured from the top
left number Yes Left, the proportion of the image width, measured from the left side
right number Yes The proportion of the width occupied on the right side of the image, measured from the left edge
usage - - - object Yes Current token usage
input_tokens - - integer Yes Billed input tokens, including image tokens
output_tokens - - integer Yes Output tokens used for answering questions. Currently not billed.

Request Example

curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/systemone' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -d '{
    "model": "MaaS_Typesafe_Jev_1.14",
    "state": {
      "subject": "Duplicate charge",
      "message": "I was charged twice for my subscription this month. Please refund the extra charge."
    },
    "questions": {
      "billing": {
        "type": "noul",
        "instructions": "Is this message about billing?"
      },
      "repeat_contact": {
        "type": "noul",
        "instructions": "Has the customer contacted support about this before?",
        "criteria": {
          "true": "Mentions a prior attempt, ticket, or that they have asked before",
          "false": "No sign of any previous contact"
        }
      },
      "tone": {
        "type": "choice",
        "instructions": "What is the tone of this message?",
        "criteria": {
          "angry": "An upset or hostile message",
          "calm": "A neutral or polite message",
          "excited": "An enthusiastic or eager message"
        }
      },
      "urgency": {
        "type": "score",
        "instructions": "How urgent is this message?",
        "criteria": [
          "Can wait",
          "Needs attention this week",
          "Needs attention today"
        ]
      }
    }
  }'

Response Example

{
  "model": "MaaS_Typesafe_Jev_1.14",
  "answers": {
    "billing": {
      "type": "noul",
      "noul": 0.98
    },
    "repeat_contact": {
      "type": "noul",
      "noul": 0.07
    },
    "tone": {
      "type": "choice",
      "choice": "calm",
      "confidence": 0.7,
      "probabilities": {
        "angry": 0.15,
        "calm": 0.8,
        "excited": 0.05
      }
    },
    "urgency": {
      "type": "score",
      "score": 1.7,
      "confidence": 0.55,
      "legend": {
        "0": "Can wait",
        "1": "Needs attention this week",
        "2": "Needs attention today"
      },
      "probabilities": {
        "0": 0.1,
        "1": 0.1,
        "2": 0.8
      }
    }
  },
  "usage": {
    "input_tokens": 120,
    "output_tokens": 12
  }
}

Box request example

Vendor Agreement Example. assetsand TypeSafe-Danger-Zone-Imagesare not currently forwarded by the MaaS gateway, so images and this header will be lost when this call is routed to the upstream service.

curl -X POST 'https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/systemone' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'TypeSafe-Danger-Zone-Images: true' \
  -d '{
    "model": "MaaS_Typesafe_Jev_1.14",
    "state": {"photo": "6e1d0b7c-kitchen"},
    "assets": {
      "6e1d0b7c-kitchen": {
        "type": "image",
        "format": "jpg",
        "data_base64": "/9j/4AAQSkZJRg..."
      }
    },
    "questions": {
      "mug_visible": {
        "type": "noul",
        "instructions": "Is there a coffee mug in `photo`?"
      },
      "mug": {
        "type": "box",
        "instructions": "the coffee mug in `photo`"
      }
    }
  }'

Box Response Example

{
  "model": "MaaS_Typesafe_Jev_1.14",
  "answers": {
    "mug_visible": {
      "type": "noul",
      "noul": 0.97
    },
    "mug": {
      "type": "box",
      "box": {
        "top": 0.25,
        "bottom": 0.82,
        "left": 0.44,
        "right": 0.84
      },
      "confidence": 0.92
    }
  },
  "usage": {
    "input_tokens": 290,
    "output_tokens": 21
  }
}

Pixel coordinates: left/rightmultiplied by the image width, top/bottommultiplied by the image height. When the target may not be in the image, use Noul in the same request to determine whether to adopt this bounding box.

SDK Calling Method

The official Python / JavaScript SDK's base_urlwill send a request to{base_url}/v1/systemone. The MaaS path is /v1/ai/{endpointPath}/systemone, and you cannot directly change the base_urlto the gateway host. The following demonstrates calling MaaS via HTTP Client.

import os
import requests

response = requests.post(
    "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/systemone",
    headers={
        "Authorization": f"Bearer {os.environ['MAAS_API_KEY']}",
        "Content-Type": "application/json",
    },
    json={
        "model": "MaaS_Typesafe_Jev_1.14",
        "state": "I was charged twice. Please help.",
        "questions": {
            "billing": {
                "type": "noul",
                "instructions": "Is this about billing?",
            },
            "tone": {
                "type": "choice",
                "instructions": "What is the tone?",
                "criteria": {"calm": None, "angry": None},
            },
        },
    },
    timeout=60,
)
response.raise_for_status()
body = response.json()
print(body["model"], body["answers"]["billing"]["noul"])
const response = await fetch(
  "https://genaiapi.cloudsway.net/v1/ai/{endpointPath}/systemone",
  {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.MAAS_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      model: "MaaS_Typesafe_Jev_1.14",
      state: "I was charged twice. Please help.",
      questions: {
        billing: {
          type: "noul",
          instructions: "Is this about billing?",
        },
        tone: {
          type: "choice",
          instructions: "What is the tone?",
          criteria: { calm: null, angry: null },
        },
      },
    }),
  },
);

const body = await response.json();
console.log(body.model, body.answers.billing.noul);