任务域决策包
日常分流不值得使用 Opus 4.8；Sonnet 4.6 是正确默认值
c
我需要分类 1 万张客服工单并按优先级路由
r
select@business_decision/customer_support_triage
C'
Sonnet 4.6 是默认选择：精度足够、成本更低、速度更快。Opus 4.8 只用于复杂升级边界情况。输入量巨大且成本优先时用 GPT-5.5。
OCG Lab任务域决策系统
D9
业务决策 / 客服工单分流
@business_decision/customer_support_triage
返回任务域
决策栈
使用
volume >1000 tickets/day, rules well-defined, cost matters
claude/sonnet-4.6使用
escalation logic is complex, ticket requires multi-step reasoning
claude/opus-4.8使用
batch input-heavy triage at scale, cost is primary constraint
openai/gpt-5.5人工门禁
high-stakes escalation decision affects customer relationship
human认知转移
模型层| ID | 关系 | 执行体 | 结果 | 输出 |
|---|---|---|---|---|
| CT-D9-001 | identify@issue_type | claude/sonnet-4.6 | 已接受 | 对定义清晰的工单类别分类能力强 |
| CT-D9-002 | escalate@condition | claude/sonnet-4.6 | 有边界 | sufficient for rule-based escalation; weaker on ambiguous multi-factor decisions |
| CT-D9-003 | escalate@condition | claude/opus-4.8 | 已接受 | stronger on complex multi-step escalation; justified cost for edge cases only |
E-D9-001有边界
Sonnet 4.6 preferred by developers over prior flagship Opus 4.5
E-D1-002E-D9-002已接受
Opus 4.8 cost is 5x higher than Sonnet 4.6 on output tokens
公开价格E-D9-003已接受
GPT-5.5 lower input cost useful for input-heavy triage
公开价格