Could you do a guide on how to select models

[BradKML]

In Claude Code Router, there are many parameters for placing different types of models https://github.com/musistudio/claude-code-router?tab=readme-ov-file#router

• default is for regular LLMs (235B or higher, can go up to 1T)
• background is for SLMs (80B or below, ideally around 30B)
• think is for reasoning models, or hybrid models in reasoning mode (more likely for LLMs rather than SLMs)
• longContext is for situations when the context gets too large, since 131K+ context is the norm, put 200K/262K/1M models here
• webSearch which people usually recommend Perplexity or Google models, alternative is :online in OpenRouter

Could LLXPRT have a sort of "router" considering that Gemini may very well be doing it "under the hood"? Is it even worth considering for multi-modal models vs thinking models vs lighter SLMs?

[acoliver]

LLxprt Code isn't a router. The intent is to add subagents and make each subagent have a different config. Presently the web search is hard coded to google but that will be configurable. So in essence it will be similar but more direct and directly controlled. Basically imagine if Claude Code supported this directly vs hacking it in some proxy...that's what we're doing.