Anthropic SDK

使用 Claude 原生格式（/v1/messages）时，推荐直接使用 Anthropic 官方 SDK，思考（thinking）、工具调用、提示词缓存等能力完整可用。令牌需要属于支持 Claude 的分组：

支持 Claude 的分组

AzClaude¥2.78/刀6 个模型

安装与初始化

Python

# pip install -U anthropic
import anthropic

client = anthropic.Anthropic(
    base_url="https://vip.zen-ai.top",  # 不要加 /v1
    api_key="sk-xxxxxxxx",
)

TypeScript

// npm install @anthropic-ai/sdk
import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://vip.zen-ai.top", // 不要加 /v1
  apiKey: process.env.VIP_API_KEY,
});

也可以设置环境变量 ANTHROPIC_BASE_URL=https://vip.zen-ai.top 与 ANTHROPIC_API_KEY=sk-xxx，代码中直接创建客户端。

基础对话

Python

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system="你是一个简洁的助手。",
    messages=[{"role": "user", "content": "用三句话介绍 Rust 语言"}],
)
print(message.content[0].text)
print(message.usage)

TypeScript

const message = await client.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  system: "你是一个简洁的助手。",
  messages: [{ role: "user", content: "用三句话介绍 Rust 语言" }],
});
for (const block of message.content) {
  if (block.type === "text") console.log(block.text);
}

流式输出

输出较长或 max_tokens 较大时建议使用流式，避免请求超时：

Python

with client.messages.stream(
    model="claude-sonnet-4-6",
    max_tokens=4096,
    messages=[{"role": "user", "content": "写一篇 500 字的科普短文：黑洞是什么"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)
    final = stream.get_final_message()
print("\n", final.usage)

TypeScript

const stream = client.messages.stream({
  model: "claude-sonnet-4-6",
  max_tokens: 4096,
  messages: [{ role: "user", content: "写一篇 500 字的科普短文：黑洞是什么" }],
});
stream.on("text", (text) => process.stdout.write(text));
const final = await stream.finalMessage();
console.log("\n", final.usage);

思考（Thinking）

Claude 4.6 及更新的模型使用自适应思考：thinking: {type: "adaptive"}，由模型自行决定思考多少，并可通过 output_config.effort 控制深度（low / medium / high / max，部分新模型还支持 xhigh）。

Python

with client.messages.stream(
    model="claude-opus-4-8",
    max_tokens=16000,
    thinking={"type": "adaptive"},
    output_config={"effort": "high"},
    messages=[{"role": "user", "content": "证明根号 2 是无理数"}],
) as stream:
    final = stream.get_final_message()

for block in final.content:
    if block.type == "thinking":
        print("【思考】", block.thinking)
    elif block.type == "text":
        print("【回答】", block.text)

TypeScript

const final = await client.messages
  .stream({
    model: "claude-opus-4-8",
    max_tokens: 16000,
    thinking: { type: "adaptive" },
    output_config: { effort: "high" },
    messages: [{ role: "user", content: "证明根号 2 是无理数" }],
  })
  .finalMessage();

for (const block of final.content) {
  if (block.type === "thinking") console.log("【思考】", block.thinking);
  else if (block.type === "text") console.log("【回答】", block.text);
}

不同模型的思考参数

• Claude 4.6 及更新模型（claude-sonnet-4-6、claude-opus-4-6 及以上）：使用 {"type": "adaptive"}。Opus 4.7 / 4.8、Opus 5 等较新模型不再接受 budget_tokens 和 temperature 等采样参数，传入会返回 400。
• Opus 4.7 及更新模型默认不返回思考文本（thinking 字段为空字符串），这不影响回答质量。
• Claude Haiku 4.5 及更早的模型：仍使用 {"type": "enabled", "budget_tokens": 8000}，budget_tokens 需小于 max_tokens。
• 多轮对话中，请把上一轮返回的 thinking 块原样带回。

工具调用

tools = [{
    "name": "get_weather",
    "description": "查询指定城市的实时天气",
    "input_schema": {
        "type": "object",
        "properties": {"city": {"type": "string", "description": "城市名称"}},
        "required": ["city"],
    },
}]
messages = [{"role": "user", "content": "北京现在天气怎么样？"}]

resp = client.messages.create(model="claude-sonnet-4-6", max_tokens=1024, tools=tools, messages=messages)

if resp.stop_reason == "tool_use":
    messages.append({"role": "assistant", "content": resp.content})
    results = []
    for block in resp.content:
        if block.type == "tool_use":
            output = f"{block.input['city']}：晴，18°C"  # 这里调用你自己的实现
            results.append({"type": "tool_result", "tool_use_id": block.id, "content": output})
    messages.append({"role": "user", "content": results})
    resp = client.messages.create(model="claude-sonnet-4-6", max_tokens=1024, tools=tools, messages=messages)

print(resp.content[-1].text)

图片与 PDF

import base64

with open("chart.png", "rb") as f:
    image_b64 = base64.standard_b64encode(f.read()).decode()
with open("report.pdf", "rb") as f:
    pdf_b64 = base64.standard_b64encode(f.read()).decode()

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=2048,
    messages=[{
        "role": "user",
        "content": [
            {"type": "image", "source": {"type": "base64", "media_type": "image/png", "data": image_b64}},
            {"type": "document", "source": {"type": "base64", "media_type": "application/pdf", "data": pdf_b64}},
            {"type": "text", "text": "结合图表和报告，总结三个关键结论。"},
        ],
    }],
)
print(message.content[0].text)

完整参数见 Claude 原生格式 → 创建消息。