Claude Prompt Caching 快取計費指南

# ❌ 錯：純字串永不快取
"content": "一大段長文..."

# ✅ 對：content block + cache_control
"content": [
    {
        "type": "text",
        "text": "一大段長文...",
        "cache_control": {"type": "ephemeral"},
    },
    {"type": "text", "text": "問題"},
]

# ❌ 錯：每換一個問題字首就變了，永遠命中不了
content = [
    {"type": "text", "text": "請回答下面問題: " + 問題},  # 這塊在變
    {"type": "text", "text": 長文, "cache_control": {"type": "ephemeral"}},
]

# ✅ 對：長文在前打標記，問題在後不打標記
content = [
    {"type": "text", "text": 長文, "cache_control": {"type": "ephemeral"}},  # 穩定
    {"type": "text", "text": 問題},                                            # 隨便變
]

import json, os, requests

URL = "https://api.apiyi.com/v1/messages"
KEY = os.environ["APIYI_API_KEY"]
HEADERS = {
    "content-type": "application/json",
    "x-api-key": KEY,
    "anthropic-version": "2023-06-01",
}

# 必須足夠長。Sonnet 4.6 至少 1024 tokens，大約 1000+ 中文字。
LONG_TEXT = open("long_document_zh.txt").read()


def ask(question: str, label: str):
    payload = {
        "model": "claude-sonnet-4-6",
        "max_tokens": 256,
        "messages": [{
            "role": "user",
            "content": [
                {"type": "text", "text": LONG_TEXT, "cache_control": {"type": "ephemeral"}},
                {"type": "text", "text": question},
            ],
        }],
    }
    r = requests.post(URL, headers=HEADERS, data=json.dumps(payload), timeout=120)
    u = r.json().get("usage", {})
    print(f"[{label}] input={u.get('input_tokens')} "
          f"write={u.get('cache_creation_input_tokens')} "
          f"read={u.get('cache_read_input_tokens')}")


ask("請概括主旨", "第1次")    # 期望 write>0, read=0
ask("請給3個關鍵詞", "第2次")  # 期望 write=0, read>0

[第1次] input=35 write=6512 read=0
[第2次] input=22 write=0    read=6512

# 每一輪請求構造時
messages[-1]["content"][-1]["cache_control"] = {"type": "ephemeral"}