Claude Code offers an
Advisor Tool
Show archive.org snapshot
which allows pairing two models.
Your agent's model will consult its advisor model to clarify its approach or to get "help" in more complex situations (like planning or debugging).
How to use
In Claude Code, type /advisor to open a chooser for the advisor model:
You may also pass the desired model as an argument to the /advisor command.
/advisor fable
Your chosen advisor model is saved to your Claude Code settings file; new sessions will use it by default.
Currently (August 2026), the advisor tool is in beta and can be enabled by setting an environment variable.
CLAUDE_CODE_ENABLE_EXPERIMENTAL_ADVISOR_TOOL=true claude
To enable it globally, populate that env variable in every terminal. For example:
# ~/.bashrc
export CLAUDE_CODE_ENABLE_EXPERIMENTAL_ADVISOR_TOOL=true
Pairings
The advisor model must be at least as potent as your regular model.
Aside from that, you can pair as you like. For Example:
- Sonnet + Opus → reasonably fast, apparently near-Opus quality.
- Sonnet + Fable → reasonably fast, pragmatic, handles tasks really well that I'd usually give to a single Opus.
- Opus + Fable → slow, but delivers strong results, and won't eat up your quota as quick as Fable.
- Sonnet + Sonnet → quick, more robust results on routine tasks, as the advisor Sonnet can catch what the regular Sonnet might miss.
The Claude Code docs suggest a few more pairings Show archive.org snapshot that I have not tried yet.
Recommendations
I've had very good results with Sonnet + Fable for mechanical tasks. My experience with Sonnet has been that it's very pragmatic, but occasionally makes mistakes or misses relevant architectural bits. A Fable advisor helps a lot with that.
When working on a more complex ticket, I usually use Opus + Fable, clarify requirements and plan the approach, and have them implement autonomously. Review churn has been lower than with a single Opus.
