Claude Haiku 4.5 API：工单分流与思考预算

curl --fail-with-body --max-time 90 https://api.apizt.com/v1/messages \
  -H "Authorization: Bearer $tokenba_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "max_tokens": 768,
    "system": "你是客服分流助手。只输出 JSON：category 取 billing、integration、incident 或 manual；priority 取 normal 或 urgent；evidence 为原文片段。不要猜测金额、账号或故障原因。",
    "messages": [{"role": "user", "content": "我充值已经到账，但 Python SDK 仍提示 401。我把 Base URL 填成了 https://api.apizt.com/v1/chat/completions，应该怎么改？"}]
  }'

import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ["tokenba_API_KEY"],
    base_url="https://api.apizt.com",
)
tickets = [
    "充值已扣款，但余额没有更新。",
    "希望把现有 OpenAI SDK 改为 apizt 地址。",
    "生产服务连续返回 503，已影响用户请求。",
]
for index, ticket in enumerate(tickets, start=1):
    result = client.messages.create(
        model="claude-haiku-4-5-20251001",
        max_tokens=512,
        system="只输出 JSON，字段为 category、priority、evidence。类别为 billing、integration、incident、manual；无法确定则 manual。",
        messages=[{"role": "user", "content": ticket}],
    )
    text = "".join(block.text for block in result.content if block.type == "text")
    print(index, text)
    print(result.usage)