- Problem statement: given a choice of multiple LLMs with different capabilities and costs, route my prompt to the best LLM
- As people started using LLMs more widely and paying for inference, smarter choices of models became attractive
- LLM routing seems to be of wide interest
- There are numerous tools available
- However, this is not close to a solved problem. The ability of LLM routers today leaves a large gap to what’s theoretically possible.