GLM
Endpoint: https://mmu.vod-qcloud.com/v1인증: VOD ApiToken / Bearer
기본 정보
| 항목 | 값 |
|---|---|
| 호출 주소 | https://mmu.vod-qcloud.com |
| 인증 | VOD 애플리케이션에서 발급한 ApiToken / Bearer |
| 출력 | 텍스트 / 구조화된 응답 |
| 응답 방식 | JSON / SSE |
지원 프로토콜
| 규격 | 경로 | 대상 |
|---|---|---|
| Chat Completions | POST /v1/chat/completions | 아래 모든 모델 |
| Messages | POST /v1/messages | 아래 모든 모델 |
아래 모델 ID에 해당 규격을 사용합니다. 인증과 공통 필드는 API Reference를 따릅니다.
샘플링 옵션
이 문서에 나열된 모든 모델에 공통으로 적용됩니다.
| 파라미터 | 설정 |
|---|---|
temperature | 설정 가능 |
top_p | 설정 가능 |
temperature는 샘플링 온도, top_p는 누적 확률 범위를 지정합니다. 한 가지 방식을 중심으로 조정합니다.
버전별 추론 설정
| 모델 ID | 설정 방식 | 비활성화 |
|---|---|---|
glm-5.1 | thinking_enabled | 비활성화 불가 |
glm-5-turbo | thinking_enabled | false |
glm-5 | thinking_enabled | false |
Chat Completions 기준입니다. 추론 내용은 reasoning_content, 최종 답변은 content에서 읽습니다. 비활성화할 수 없는 버전에는 비활성화 옵션을 전달하지 않습니다.
요청 파라미터
| 파라미터 | 필수 | 타입 | 설명 |
|---|---|---|---|
model | 필수 | String | 버전별 옵션 표의 모델 ID |
messages | 필수 | Array | Chat Completions / Messages의 대화 입력 |
max_tokens | 선택 | Integer | Chat Completions / Messages의 생성 한도. Messages에서는 필수 |
stream | 선택 | Boolean | true: SSE / false: JSON |
요청 예시
Chat Completions
json
{
"model": "glm-5.1",
"messages": [
{
"role": "user",
"content": "Reply with exactly OK and nothing else."
}
],
"max_tokens": 512,
"stream": false
}응답
json
{
"id": "01a0de30-62c1-759a-a272-4c9750a9cebc",
"object": "chat.completion",
"model": "glm-5.1",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "OK",
"reasoning_content": "1. **Analyze the Request:**\n * Input: \"Reply with exactly OK and nothing else.\"\n * Constraint 1: Reply with \"OK\".\n * Constraint 2: Exactly that word, nothing else (no punctuation, no extra words, no formatting).\n\n2. **Formulate the Output:**\n * Target string: \"OK\"\n\n3. **Verify against constraints:**\n * Is it exactly \"OK\"? Yes.\n * Is there anything else? No.\n\n4. **Final Output Generation:**\n * OK"
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 13,
"prompt_tokens_details": {
"audio_tokens": 0,
"cached_tokens": 0
},
"prompt_cached_tokens_details": {
"audio_tokens": 0
},
"completion_tokens": 125,
"completion_tokens_details": {
"accepted_prediction_tokens": 0,
"audio_tokens": 0,
"reasoning_tokens": 0,
"rejected_prediction_tokens": 0
},
"total_tokens": 138
}
}Messages
json
{
"model": "glm-5.1",
"messages": [
{
"role": "user",
"content": "Reply with exactly OK and nothing else."
}
],
"max_tokens": 512,
"stream": false
}응답
json
{
"id": "01a0de33-76c0-751e-9056-bd61bd234513",
"model": "glm-5.1",
"type": "message",
"role": "assistant",
"content": [
{
"thinking": "1. **Analyze the Request:** The user is asking for a specific response: \"OK\" and nothing else.\n2. **Formulate the Output:** The output must strictly be the string \"OK\" without any additional text, punctuation (beyond what constitutes \"OK\"), or newlines if possible (though a standard single newline at the end is usually unavoidable/system-level, the raw text should just be \"OK\").\n3. **Verify Constraints:** Exactly \"OK\". No \"Sure\", no \"OK.\", no explanation.\n4. **Final Output Generation:** OK",
"type": "thinking"
},
{
"text": "OK",
"type": "text"
}
],
"stop_reason": "end_turn",
"usage": {
"cache_creation_input_tokens": 0,
"cache_read_input_tokens": 0,
"input_tokens": 13,
"output_tokens": 121
}
}특수 설정
대화 이력은 messages에 순서대로 전달합니다. 추론을 반환하는 모델에서는 reasoning_content를 최종 답변과 분리합니다. 생성 한도에는 추론 토큰도 포함되므로 긴 추론 작업에는 충분한 max_tokens를 지정합니다.