LLM Routing

LLM routing means directing a coding task to the model best suited for it. Some tasks need a massive context window. Others need fast, cheap responses. Routing is how you match the task to the right provider.

How claude-multi handles it

claude-multi doesn’t do automatic routing. Each instance is pinned to a single provider. But because instances are cheap to create and independent, you can run multiple instances side by side and pick the right one for the job.

Common routing patterns:

• Use a GLM instance for general coding (fast, cost-effective)
• Use a MiniMax instance when you need a huge context window (1M tokens)
• Use a DeepSeek instance for deep reasoning tasks
• Use an Anthropic instance when you need the original Claude experience

Why not automatic routing

Automatic routing adds latency, complexity, and a failure point. claude-multi keeps it simple: you create instances for the providers you want, and you pick which one to use. No proxy, no daemon, no middleware deciding for you.

Practical tip

Create a naming convention that makes routing obvious. claude-glm for fast work, claude-minimax for large files, claude-deepseek for complex debugging. The wrapper scripts make this natural.