Free guide
How Claude Code's Advisor Works (and How to Turn It On)
The trick going around X is a real Claude Code feature: a cheaper model does the work and asks a stronger model for advice only at the hard moments. Here's how the handoff works and how to set it up.

The short version
- Your main model runs the whole job. The advisor only steps in when Claude decides it needs a second opinion.
- The advisor reads the whole conversation, every tool call and result, and sends back a plan, a correction or a stop. It never runs tools itself.
- Advisor calls are billed at the advisor model's rates on top of your main model, but Anthropic says pairing a faster model with a stronger advisor typically costs less than running the stronger model throughout.
How it actually works
Think of a caver with a phone line to base camp
The caver does all the climbing. Base camp never goes underground. When the caver hits a fork they can't judge, they ring up, base camp reads the whole logbook and sends down a short note. In Claude Code, your main model is the caver and the advisor is base camp.
The main model does the work
Anthropic calls the main model the executor. It runs the task end to end: it calls tools, reads files, runs tests and keeps going. Most turns never touch the advisor, so most of the run stays at the main model's price.
Claude decides when to ring
There's no rule that fires the advisor. Claude chooses, and it tends to call it before committing to an approach, when an error keeps coming back, and before declaring a task done. You can also ask for it in your prompt, for example: consult the advisor before you continue.
The advisor reads everything and sends a note
When Claude calls the advisor, Anthropic's servers run the advisor model on the full conversation, including your system prompt, the tool definitions and every tool call and result. The advisor returns guidance, typically a few hundred words: a plan, a correction or a signal to stop. It never calls tools and never writes the code. Claude reads the note and carries on, and if the note clashes with what it can see in the files, it says so instead of following it blindly.
How to set it up
1.Check your setup
Update Claude Code first. The advisor works on the Anthropic API, Claude Platform on AWS or Claude Models on Google Cloud (Anthropic managed), on subscription plans and API billing. It isn't available on Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry.
claude update2.Pick a main model, then an advisor
Anthropic's recommended setup is Sonnet as the main model with Opus as the advisor. The advisor has to rank at or above your main model, so an Opus main model won't take a Sonnet advisor.
/model sonnet /advisor opus3.Or set it once in your settings
To make it your default without opening a session, add advisorModel to your settings file. To try it for one session only, launch with the flag instead.
{ "advisorModel": "opus" } claude --advisor opus4.Give it a task worth advising on
The advisor pays off on long jobs where most turns are routine but the plan decides the outcome: a big refactor, a bug that keeps coming back, or work you want checked before it's called done. When it runs you'll see an Advising line; press Ctrl+O to read the advice.
Refactor the payment module to use the new API client. Consult the advisor before you pick an approach and before you say it's done.5.Watch the cost, and turn it off if it doesn't help
Check /usage after a few tasks. Advisor tokens count toward your session totals. If short tasks aren't getting better, switch it off.
/usage /advisor off
The fine print
- The advisor is experimental. Anthropic says its behavior, pricing and availability may change.
- Every advisor call reads the whole conversation again at the advisor model's rates, and that read isn't cached, so long sessions with many calls cost more.
- Anthropic's cost results come from its own April 2026 tests with Sonnet 4.6 and Opus 4.6, for example Terminal-Bench 2.0 at $0.88 per task with the advisor vs $0.94 for Sonnet alone. Your results depend on your tasks.
- On some plans Fable as the advisor bills to usage credits and needs a one-time consent through /model fable first.
- The viral posts about a four-model Fable, Sonnet, Haiku and Opus stack and a 3.4x saving aren't from Anthropic. We couldn't find an Anthropic source for them.
FAQ
What does /advisor do in Claude Code?
It sets a second, usually stronger model that your main model can consult mid-task. Run /advisor to open a picker, /advisor opus to set Opus directly, or /advisor off to turn it off. The choice is saved as your default.
Does the advisor write code or run commands?
No. The advisor reads the conversation and returns guidance. Anthropic says it never calls tools or produces user-facing output. Your main model does all the work.
Is the advisor cheaper than just using Opus?
Usually, because the stronger model only runs at decision points instead of every turn. In Anthropic's tests, Sonnet with an Opus advisor scored higher than Sonnet alone and cost less per task. Each advisor call still adds tokens at the advisor's rates.
When does Claude call the advisor?
Claude decides. It tends to call it before committing to an approach, when an error keeps recurring and before declaring a task done. You can ask for it in your prompt, but there's no setting to force or cap calls in Claude Code.
Which models can be the advisor?
The advisor must rank at or above your main model. With Sonnet 5.5 as the main model, Claude Code accepts Fable, Opus 5 or later, or Sonnet 5.5 as the advisor. Haiku 4.5 can call the advisor but can't act as one.
Sources
- Escalate hard decisions with the advisor tool (Claude Code docs)
- The advisor strategy (Anthropic blog, Apr 9 2026)
- Advisor tool (Claude API docs)
2026-11-15: against the Claude Code advisor docs, Anthropic's advisor strategy post and the Claude API advisor tool docs, plus the /advisor picker in Claude Code 2.1.312.
Want help building with tools like this?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolMore free guides
Qwen4-Max: Why Alibaba's New Model Costs a Third as Much per Job
Alibaba's full Qwen4-Max release scores higher than the model before it and, in Artificial Analysis' test, costs about a third as much per task. Here's where the savings come from, how to try it today, and where it still trails.
How ChatGPT Wallet Pays Without Seeing Your Card (and How to Set It Up)
OpenAI's new ChatGPT Wallet lets your dot pay for things on your behalf, and neither the agent nor the store ever sees your real card number. Here's how the stand-in card works and how to set it up.
Claude Code Custom Themes: Animated Backdrops and Your Own Colors
The Claude Code desktop app can now wear a theme: your own colors and an animated scene behind the chat. Here is how a theme works, two ways to get one today, and why you should treat a theme like code.