Developer Guide

The Quality Estimation (QE) Service provides automated translation quality scores normalized to a 0–100 scale.

Base URL:

https://qe.service.console.custom.mt

1. Authentication

Each request must include an API key.

Get the key in Console http://Custom.MT :
Translations > Connections
Console: https://console.custom.mt/

Header:

API-KEY: <your_api_key>

2. Request Structure

Endpoint:

POST /qe

Headers:

accept: application/json
Content-Type: application/json
API-KEY: <your_api_key>

Body Example:

{
  "messages": [
    {
      "source": "Hallo Welt!",
      "target": "Hello world!"
    }
  ],
  "language_pair": {
    "source_language": "de",
    "target_language": "en"
  },
  "models": [
    "claude"
  ]
}

3. Response Structure

Example:

{
  "messages": [
    {
      "score": {
        "claude": {
          "score": 95,
          "usage": {
            "prompt_tokens": 63,
            "completion_tokens": 5
          },
          "prompt_error": null,
          "error": null
        }
      },
      "message": {
        "source": "Hallo Welt!",
        "target": "Hello world!",
        "source_language": "de",
        "target_language": "en"
      }
    }
  ],
  "usage": {
    "claude-3-7-sonnet-20250219": {
      "prompt_tokens": 63,
      "completion_tokens": 5
    }
  }
}

Key fields:

score → integer 0–100 (normalized quality)
usage → token usage (for LLM-based models)
error / prompt_error → null when successful

4. Supported Models

Model	Description

Model	Description
`mmt`	ModernMT QE scoring
`widn`	Widn (Unbabel) QE
`taus`	TAUS QE API
`open_ai`	OpenAI LLM-based QE
`claude`	Claude (Anthropic) QE

All models return scores normalized to 0–100.

5. Example Curl

curl -X POST \
  'https://qe.service.console.custom.mt/qe' \
  -H 'accept: application/json' \
  -H 'API-KEY: <your_api_key>' \
  -H 'Content-Type: application/json' \
  -d '{
    "messages": [
      {
        "source": "Hallo Welt!",
        "target": "Hello world!"
      }
    ],
    "language_pair": {
      "source_language": "de",
      "target_language": "en"
    },
    "models": [
      "claude"
    ]
  }'

6. API Reference

Full interactive API documentation is available here:
QE Service Swagger UI

You can explore endpoints, test requests, and generate code samples for different languages directly from this page.