Claude 协议：Anthropic Messages

POST /v1/messages 是 Anthropic 原生格式：system 是顶层字段，content 用 block 数组表达。

最小请求

curl "$BASE_URL/v1/messages" \
  -H "x-api-key: $NEW_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<模型ID>",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "你好"}]
  }'

Authorization: Bearer $NEW_API_KEY 同样可用，两种头二选一。

与 OpenAI 格式的差异

项目	OpenAI Chat	Anthropic Messages
系统提示	messages 里 role: system	顶层 system 字段
输出上限	max_tokens 可选	max_tokens 必填
内容格式	content 字符串	字符串或 block 数组
必带头	无	anthropic-version
停止原因	finish_reason	stop_reason

内容块

{
  "role": "user",
  "content": [
    { "type": "text", "text": "这张图里是什么？" },
    { "type": "image", "source": { "type": "base64", "media_type": "image/png", "data": "..." } }
  ]
}

SDK

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.nowhub.org",   # 不要带 /v1
    api_key="sk-...",
)

msg = client.messages.create(
    model="<模型ID>",
    max_tokens=1024,
    system="你是一个简洁的助手",
    messages=[{"role": "user", "content": "你好"}],
)
print(msg.content[0].text)

流式

with client.messages.stream(
    model="<模型ID>",
    max_tokens=1024,
    messages=[{"role": "user", "content": "写一段 100 字的介绍"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

响应要点

{
  "type": "message",
  "role": "assistant",
  "content": [{ "type": "text", "text": "..." }],
  "stop_reason": "end_turn",
  "usage": { "input_tokens": 12, "output_tokens": 5 }
}

stop_reason 为 max_tokens 表示截断；usage 中的 cache_read_input_tokens / cache_creation_input_tokens 表示命中或写入提示词缓存，按对应单价计费。