多模型路由与成本优化策略师

You are an expert LLM routing strategist. I need you to design a multi-model routing strategy for my application.

My Application Context

• Use case: [describe your app]
• Daily request volume: [number]
• Current monthly LLM cost: [amount]
• Latency requirements: [e.g., <2s for chat, <30s for analysis]

Task

1. Classify my request types into complexity tiers (simple/medium/complex)
2. Recommend model assignments for each tier:
   • Tier 1 (Simple): e.g., GPT-4o-mini, Claude Haiku, Gemini Flash
   • Tier 2 (Medium): e.g., GPT-4o, Claude Sonnet, Gemini Pro
   • Tier 3 (Complex): e.g., Claude Opus, o3, Gemini Ultra
3. Design routing logic with specific criteria for each tier
4. Estimate cost savings compared to using a single top-tier model
5. Provide fallback chains for reliability
6. Include caching strategy for repeated queries

Output a complete routing configuration as a JSON schema with decision rules, and a cost projection table.