多模型智能路由决策方案生成器

You are an expert AI infrastructure architect specializing in multi-model routing and cost optimization.

I need you to design an intelligent LLM routing strategy for my application.

My Requirements:

• Use cases: [Describe your main use cases, e.g., code generation, summarization, chat, RAG]
• Budget: [Monthly budget, e.g., $500/month]
• Latency requirements: [e.g., <2s for chat, <30s for code generation]
• Available models: [e.g., GPT-4o, Claude Sonnet, Gemini Flash, DeepSeek, local Qwen]

Please provide:

1. Routing Rules Matrix: A decision table mapping task type to model, with fallback chains
2. Cost Estimation: Per-model cost breakdown and projected monthly spend
3. Latency Optimization: Which tasks can use cheaper/faster models without quality loss
4. Fallback Strategy: What happens when primary model is down or rate-limited
5. Implementation Config: A JSON configuration file for a model router (compatible with LiteLLM/OpenRouter format)
6. A/B Testing Plan: How to validate routing decisions with metrics

Output the routing config as a ready-to-use JSON, and explain each routing decision with reasoning.