Token Counting & Cost Optimization

来源: Claude Token Counting 官方文档 学习日期: 2026-04-12

概述

Token Counting 让你在发送请求前预先计算 token 数量，帮助：

• 主动管理速率限制和成本
• 智能路由模型决策
• 优化 prompt 到特定长度

注意：Token 计数为估算值，实际用量可能略有差异。系统自动添加的 token 不计入账单。

---

API 端点

POST https://api.anthropic.com/v1/messages/count_tokens

支持所有 active models。

---

基本用法

Python

import anthropic

client = anthropic.Anthropic()

response = client.messages.count_tokens(
    model="claude-opus-4-6",
    system="You are a scientist",
    messages=[{"role": "user", "content": "Hello, Claude"}],
)
print(response.json())
# {"input_tokens": 14}

TypeScript

import Anthropic from "@anthropic-ai/sdk";
const client = new Anthropic();

const response = await client.messages.countTokens({
  model: "claude-opus-4-6",
  system: "You are a scientist",
  messages: [{ role: "user", content: "Hello, Claude" }],
});
console.log(response);

CLI

ant messages count-tokens \
  --model claude-opus-4-6 \
  --system "You are a scientist" \
  --message '{role: user, content: "Hello, Claude"}'

---

工具调用场景

计算含 Tool Use 的 token 时，schema 定义也会消耗 token。

response = client.messages.count_tokens(
    model="claude-opus-4-6",
    tools=[{
        "name": "get_weather",
        "description": "Get the current weather in a given location",
        "input_schema": {
            "type": "object",
            "properties": {
                "location": {"type": "string", "description": "The city and state"}
            },
            "required": ["location"]
        }
    }],
    messages=[{"role": "user", "content": "What's the weather like in San Francisco?"}],
)

---

成本优化策略

1. 预先计算预算

MAX_TOKENS = 100000  # $0.015/opus-4-6 per 1K tokens

def estimate_cost(prompt_tokens, model="claude-opus-4-6"):
    RATES = {
        "claude-opus-4-6": {"input": 0.015, "output": 0.075},
        "claude-sonnet-4-6": {"input": 0.003, "output": 0.015},
    }
    # ... calculate

2. 智能路由

短 prompt (< 4K tokens)	长 prompt (> 4K tokens)
Claude Sonnet (便宜 5x)	Claude Opus (更强能力)

3. Prompt 压缩

• 移除冗余示例
• 使用系统 prompt 复用通用指令
• 截断历史消息（保留最近 N 轮）

---

与现有文档的关系

• Cost Optimization - 成本优化进阶指南
• Local Embedding - 本地 embedding 减少 API 成本