一份文档，
三套协议。

# 方式一 · OpenAI 风格（推荐）
Authorization: Bearer sk-你的令牌

# 方式二 · Anthropic 风格（/v1/messages 也接受方式一）
x-api-key: sk-你的令牌
anthropic-version: 2023-06-01

{
  "model": "qwen3.8-max",
  "messages": [{
    "role": "user",
    "content": [
      { "type": "text", "text": "这张图里有什么？" },
      { "type": "image_url",
        "image_url": { "url": "https://example.com/photo.jpg", "detail": "auto" } }
    ]
  }]
}

{
  "id": "chatcmpl-9f3a…",
  "object": "chat.completion",
  "created": 1756900000,
  "model": "glm-5.3",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "我是一个简洁的 AI 助手。" },
    "finish_reason": "stop"   // stop | length | tool_calls | content_filter
  }],
  "usage": {
    "prompt_tokens": 28,
    "completion_tokens": 12,
    "total_tokens": 40
  }
}

resp = client.responses.create(
    model="gpt-5.6-luna",
    instructions="你是一个简洁的助手。",      # 等价于 system prompt
    input="用一句话介绍你自己",             # 字符串或 message 数组
    max_output_tokens=512,
    reasoning={"effort": "medium"},        # 推理模型可选
)
print(resp.output_text)                   # SDK 便捷字段, 拼接全部文本输出

{
  "id": "msg_01Xa…",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet-5",
  "content": [
    { "type": "text", "text": "我是一个简洁的 AI 助手。" }
  ],
  "stop_reason": "end_turn",   // end_turn | max_tokens | stop_sequence | tool_use
  "stop_sequence": null,
  "usage": { "input_tokens": 31, "output_tokens": 14 }
}

img = client.images.generate(
    model="gpt-image-2",
    prompt="极简风格的红色柠檬 logo，白底，矢量感",
    size="1024x1024",        # 1024x1024 | 1536x1024 | 1024x1536
    quality="medium",        # low | medium | high
    n=1,
)
print(img.data[0].url or img.data[0].b64_json)

emb = client.embeddings.create(
    model="你的嵌入模型",
    input=["第一段文本", "第二段文本"],   # 字符串或字符串数组, 批量更省
    encoding_format="float",             # float | base64
)
vec = emb.data[0].embedding                 # list[float]
print(len(vec), emb.usage.total_tokens)

"response_format": {
  "type": "json_schema",
  "json_schema": {
    "name": "sentiment",
    "strict": true,
    "schema": {
      "type": "object",
      "properties": {
        "label": { "type": "string", "enum": ["positive", "neutral", "negative"] },
        "score": { "type": "number" }
      },
      "required": ["label", "score"],
      "additionalProperties": false
    }
  }
}

data: {"id":"chatcmpl-…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"role":"assistant","content":""},"finish_reason":null}]}

data: {"id":"chatcmpl-…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"我是"},"finish_reason":null}]}

data: {"id":"chatcmpl-…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"一个助手。"},"finish_reason":null}]}

data: {"id":"chatcmpl-…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: {"id":"chatcmpl-…","choices":[],"usage":{"prompt_tokens":28,"completion_tokens":12,"total_tokens":40}}   ← 需 stream_options.include_usage

data: [DONE]

with client.messages.stream(
    model="claude-sonnet-5", max_tokens=1024,
    messages=[{"role": "user", "content": "写一首关于柠檬的短诗"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)
final = stream.get_final_message()
print(final.usage)

import json

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "查询指定城市的实时天气",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string", "description": "城市名"}},
            "required": ["city"],
        },
    },
}]
messages = [{"role": "user", "content": "上海现在多少度？"}]

# 第一轮: 模型决定调用工具
r1 = client.chat.completions.create(model="glm-5.3", messages=messages, tools=tools)
call = r1.choices[0].message.tool_calls[0]
args = json.loads(call.function.arguments)        # {"city": "上海"}

# 你执行真正的函数
result = get_weather(**args)                       # 例: {"temp": 27, "desc": "多云"}

# 第二轮: 回填结果, 模型组织答案
messages += [
    r1.choices[0].message,                          # 含 tool_calls 的 assistant 消息
    {"role": "tool", "tool_call_id": call.id, "content": json.dumps(result)},
]
r2 = client.chat.completions.create(model="glm-5.3", messages=messages, tools=tools)
print(r2.choices[0].message.content)           # "上海现在 27℃，多云。"

tools = [{
    "name": "get_weather",
    "description": "查询指定城市的实时天气",
    "input_schema": {"type": "object",
                     "properties": {"city": {"type": "string"}},
                     "required": ["city"]},
}]
messages = [{"role": "user", "content": "上海现在多少度？"}]

r1 = client.messages.create(model="claude-sonnet-5", max_tokens=1024,
                            tools=tools, messages=messages)
use = next(b for b in r1.content if b.type == "tool_use")
result = get_weather(**use.input)

messages += [
    {"role": "assistant", "content": r1.content},
    {"role": "user", "content": [{
        "type": "tool_result",
        "tool_use_id": use.id,
        "content": json.dumps(result),
    }]},
]
r2 = client.messages.create(model="claude-sonnet-5", max_tokens=1024,
                            tools=tools, messages=messages)
print(r2.content[0].text)

import time
from openai import OpenAI, RateLimitError, APIStatusError

def ask(messages, retries=4):
    for i in range(retries):
        try:
            return client.chat.completions.create(model="glm-5.3", messages=messages)
        except RateLimitError:
            time.sleep(2 ** i)                       # 1s, 2s, 4s, 8s
        except APIStatusError as e:
            if e.status_code in (502, 503, 529):
                time.sleep(2 ** i)
            else:
                raise                                # 4xx 参数问题不重试
    raise RuntimeError("retries exhausted")

export ANTHROPIC_BASE_URL="https://ai.lemongir.shop"
export ANTHROPIC_API_KEY="sk-你的令牌"
export ANTHROPIC_MODEL="claude-sonnet-5"          # 或任何文本模型, 如 kimi-k2.7-code
claude

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="glm-5.3",
    api_key="sk-你的令牌",
    base_url="https://ai.lemongir.shop/v1",
    temperature=0.3,
)
print(llm.invoke("你好").content)