google-gemini/gemini-cli

More configurable numeric routing

Open

#21805 opened on Mar 10, 2026

View on GitHub
 (6 comments) (0 reactions) (1 assignee)TypeScript (103,992 stars) (13,657 forks)batch import
aiq/routingarea/agenthelp wantedkind/enhancementkind/featurepriority/p2status/bot-triaged

Description

What would you like to be added?

I would like to have much more control over routing. For numerical routing, this would mean controlling what levels get sent to which models, for example something like:

  • <10 -> local gemma
  • <30 -> 2.5 flash
  • <80 -> 3 flash
  • >80 -> 3 pro

Why is this needed?

Currently, the only routing available is to choose between the flash and pro models for the current major version. This is quite restrictive, especially as even 3 flash is quite capable (and quite quota- and rate-limited).

Additional context

No response

Contributor guide