OpenAI プロンプトキャッシュ課金ガイド

# ❌ Wrong: dynamic content at the start of system — prefix changes every time, never hits
messages = [
    {"role": "system", "content": f"Current time {datetime.now()}. You are an assistant." + long_instructions},
    {"role": "user", "content": question},
]

# ✅ Right: long instructions and tool definitions stay stable up front; dynamic bits go last
messages = [
    {"role": "system", "content": long_instructions},          # stable — will hit
    {"role": "user", "content": f"Current time {datetime.now()}. {question}"},  # volatile — last
]

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APIYI_API_KEY"],
    base_url="https://api.apiyi.com/v1"
)

# Must be long enough: at least 1024 tokens (~750+ English words)
LONG_SYSTEM = open("long_instructions.txt").read()


def ask(question: str, label: str):
    r = client.chat.completions.create(
        model="gpt-5.4",
        messages=[
            {"role": "system", "content": LONG_SYSTEM},
            {"role": "user", "content": question},
        ],
    )
    cached = r.usage.prompt_tokens_details.cached_tokens
    print(f"[{label}] input={r.usage.prompt_tokens} cached={cached}")


ask("Summarize the key points", "1st")   # expect cached=0
ask("Give 3 keywords", "2nd")            # expect cached ≈ prefix length

[1st] input=2330 cached=0
[2nd] input=2335 cached=2304

r = client.chat.completions.create(
    model="gpt-5.4",
    messages=messages,
    prompt_cache_key="user-12345"  # pin routing per user/session
)