Anthropic 消息

通过 CometAPI 使用 Anthropic Messages API，访问支持扩展思考、Prompt 缓存、工具使用、网页搜索/抓取、流式输出以及 effort control 的 Claude 模型。

POST /v1/messages

Python

Basic

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.cometapi.com",
    api_key="<COMETAPI_KEY>",
)

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system="You are a helpful assistant.",
    messages=[
        {"role": "user", "content": "Hello, world"}
    ],
)

print(message.content[0].text)

200

{
  "id": "<string>",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "<string>",
      "thinking": "<string>",
      "signature": "<string>",
      "id": "<string>",
      "name": "<string>",
      "input": {}
    }
  ],
  "model": "<string>",
  "stop_reason": "end_turn",
  "stop_sequence": "<string>",
  "usage": {
    "input_tokens": 123,
    "output_tokens": 123,
    "cache_creation_input_tokens": 123,
    "cache_read_input_tokens": 123,
    "cache_creation": {
      "ephemeral_5m_input_tokens": 123,
      "ephemeral_1h_input_tokens": 123
    }
  }
}

​

概述

CometAPI 原生支持 Anthropic Messages API，让你可以直接访问具备全部 Anthropic 专属功能的 Claude 模型。对于扩展思考、Prompt 缓存和 effort control 等 Claude 独有能力，请使用此端点。

​

快速开始

使用官方 Anthropic SDK——只需将 base URL 设置为 CometAPI：

Python

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.cometapi.com",
    api_key="<COMETAPI_KEY>",
)

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)
print(message.content[0].text)

支持使用 x-api-key 和 Authorization: Bearer 请求头进行身份验证。官方 Anthropic SDK 默认使用 x-api-key。

​

扩展思考

使用 thinking 参数启用 Claude 的分步推理。响应中会包含 thinking 内容块，在最终答案前展示 Claude 的内部推理过程。

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=16000,
    thinking={
        "type": "enabled",
        "budget_tokens": 10000,
    },
    messages=[
        {"role": "user", "content": "Prove that there are infinitely many primes."}
    ],
)

for block in message.content:
    if block.type == "thinking":
        print(f"Thinking: {block.thinking[:200]}...")
    elif block.type == "text":
        print(f"Answer: {block.text}")

Thinking 要求 budget_tokens 的最小值为 1,024。Thinking 所消耗的 Token 会计入你的 max_tokens 限制，因此请将 max_tokens 设置得足够高，以同时容纳 thinking 和响应内容。

---

​

Prompt 缓存

缓存较大的 system prompt 或对话前缀，以降低后续请求的延迟和成本。为需要缓存的内容块添加 cache_control：

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system=[
        {
            "type": "text",
            "text": "You are an expert code reviewer. [Long detailed instructions...]",
            "cache_control": {"type": "ephemeral"},
        }
    ],
    messages=[{"role": "user", "content": "Review this code..."}],
)

缓存使用情况会在响应的 usage 字段中返回：

• cache_creation_input_tokens — 写入缓存的 tokens（按更高费率计费）
• cache_read_input_tokens — 从缓存读取的 tokens（按较低费率计费）

Prompt 缓存要求被缓存的内容块至少包含 1,024 tokens。短于此长度的内容将不会被缓存。

---

​

流式输出（Streaming）

通过设置 stream: true，使用 Server-Sent Events (SSE) 进行流式响应。事件会按以下顺序到达：

1. message_start — 包含消息元数据和初始 usage
2. content_block_start — 标记每个内容块的开始
3. content_block_delta — 增量文本片段（text_delta）
4. content_block_stop — 标记每个内容块的结束
5. message_delta — 最终的 stop_reason 和完整的 usage
6. message_stop — 表示流结束

with client.messages.stream(
    model="claude-sonnet-4-6",
    max_tokens=256,
    messages=[{"role": "user", "content": "Hello"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="")

---

​

努力程度控制

使用 output_config.effort 控制 Claude 在生成响应时投入多少计算努力：

message = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Summarize this briefly."}
    ],
    output_config={"effort": "low"},  # "low", "medium", or "high"
)

---

​

服务器工具

Claude 支持运行在 Anthropic 基础设施上的服务端工具：

Web Fetch

从 URL 抓取并分析内容：

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Analyze the content at https://arxiv.org/abs/1512.03385"}
    ],
    tools=[
        {"type": "web_fetch_20250910", "name": "web_fetch", "max_uses": 5}
    ],
)

Web Search

搜索网络以获取实时信息：

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "What are the latest developments in AI?"}
    ],
    tools=[
        {"type": "web_search_20250305", "name": "web_search", "max_uses": 5}
    ],
)

---

​

响应示例

来自 CometAPI 的 Anthropic 端点的典型响应：

{
  "id": "msg_bdrk_01UjHdmSztrL7QYYm7CKBDFB",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "Hello!"
    }
  ],
  "model": "claude-sonnet-4-6",
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 19,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 0
    },
    "output_tokens": 4
  }
}

---

​

与 OpenAI-Compatible 端点的关键区别

功能	Anthropic Messages (/v1/messages)	OpenAI-Compatible (/v1/chat/completions)
扩展思考	带有 budget_tokens 的 thinking 参数	不可用
Prompt 缓存	内容块上的 cache_control	不可用
努力程度控制	output_config.effort	不可用
Web 抓取/搜索	服务器工具（web_fetch, web_search）	不可用
认证请求头	x-api-key 或 Bearer	仅 Bearer
响应格式	Anthropic 格式（content 块）	OpenAI 格式（choices, message）
模型	仅 Claude	多提供商（GPT、Claude、Gemini 等）

授权

x-api-key · string · header · 必填

Your CometAPI key passed via the x-api-key header. Authorization: Bearer <key> is also supported.

请求头

anthropic-version · string · 默认值: 2023-06-01

The Anthropic API version to use. Defaults to 2023-06-01.

示例:

"2023-06-01"

anthropic-beta · string

Comma-separated list of beta features to enable. Examples: max-tokens-3-5-sonnet-2024-07-15, pdfs-2024-09-25, output-128k-2025-02-19.

请求体

application/json

model · string · 必填

The Claude model to use. See the Models page for current Claude model IDs.

示例:

"claude-sonnet-4-6"

messages · object[] · 必填

The conversation messages. Must alternate between user and assistant roles. Each message's content can be a string or an array of content blocks (text, image, document, tool_use, tool_result). There is a limit of 100,000 messages per request.

Show child attributes

max_tokens · integer · 必填

The maximum number of tokens to generate. The model may stop before reaching this limit. When using thinking, the thinking tokens count towards this limit.

必填范围: x >= 1

示例:

1024

system · string object[]

System prompt providing context and instructions to Claude. Can be a plain string or an array of content blocks (useful for prompt caching).

temperature · number · 默认值: 1

Controls randomness in the response. Range: 0.0–1.0. Use lower values for analytical tasks and higher values for creative tasks. Defaults to 1.0.

必填范围: 0 <= x <= 1

top_p · number

Nucleus sampling threshold. Only tokens with cumulative probability up to this value are considered. Range: 0.0–1.0. Use either temperature or top_p, not both.

必填范围: 0 <= x <= 1

top_k · integer

Only sample from the top K most probable tokens. Recommended for advanced use cases only.

必填范围: x >= 0

stream · boolean · 默认值: false

If true, stream the response incrementally using Server-Sent Events (SSE). Events include message_start, content_block_start, content_block_delta, content_block_stop, message_delta, and message_stop.

stop_sequences · string[]

Custom strings that cause the model to stop generating when encountered. The stop sequence is not included in the response.

thinking · object

Enable extended thinking — Claude's step-by-step reasoning process. When enabled, the response includes thinking content blocks before the answer. Requires a minimum budget_tokens of 1,024.

Show child attributes

tools · object[]

Tools the model may use. Supports client-defined functions, web search (web_search_20250305), web fetch (web_fetch_20250910), code execution (code_execution_20250522), and more.

Show child attributes

tool_choice · object

Controls how the model uses tools.

Show child attributes

metadata · object

Request metadata for tracking and analytics.

Show child attributes

output_config · object

Configuration for output behavior.

Show child attributes

service_tier · enum<string>

The service tier to use. auto tries priority capacity first, standard_only uses only standard capacity.

可用选项: auto, standard_only

响应

200 - application/json

Successful response. When stream is true, the response is a stream of SSE events.

id · string

Unique identifier for this message (e.g., msg_01XFDUDYJgAACzvnptvVoYEL).

type · enum<string>

Always message.

可用选项: message

role · enum<string>

Always assistant.

可用选项: assistant

content · object[]

The response content blocks. May include text, thinking, tool_use, and other block types.

Show child attributes

model · string

The specific model version that generated this response (e.g., claude-sonnet-4-6).

stop_reason · enum<string>

Why the model stopped generating.

可用选项: end_turn, max_tokens, stop_sequence, tool_use, pause_turn

stop_sequence · string | null

The stop sequence that caused the model to stop, if applicable.

usage · object

Token usage statistics.

Show child attributes