How to stop wasting your Claude limit: pick the right model and use /advisor (step by step)
Claude has four models and most people run one on everything. A step-by-step guide to matching each job to Haiku 5.5, Sonnet 5.5, Opus 5.5 or Fable 5.2, and using the /advisor command without burning 8x the usage.
Claude has four models, and most people run one of them on everything. That is the quickest way to hit your usage limit. This guide shows you how to match each job to the right model, and how to use the /advisor command so the smartest model only steps in when it is actually needed. Every step below is something I ran myself in Claude Code.
The short version
- Use Sonnet 5.5 for most of your day.
- Switch to Haiku 5.5 for quick questions and small edits, and to Opus 5.5 for builds that have to be right first time.
- Turn on Fable 5.2 as your advisor for long or tricky jobs, so it only gets called when Claude is stuck.
- Turn the advisor off for small jobs. Leaving it on can cost far more than it saves.
The cheat sheet
Save this table. The usage column comes from my own test further down, measured against Sonnet.
| Model | Use it for | Usage vs Sonnet | Type this |
|---|---|---|---|
| Haiku 5.5 | Quick questions, small edits | A small fraction | /model haiku |
| Sonnet 5.5 | Daily fixes and edits | 1x | /model sonnet |
| Opus 5.5 | Builds that must be right first time | About 2x | /model opus |
| Fable 5.2 | Only when you are stuck | About 4x | /advisor fable |
| Small job | Advisor off | Saves the most | /advisor off |
Before you start
- Claude Code installed. Everything here happens inside Claude Code, in your terminal or the Claude desktop app.
- Version 2.1.321 or later. Fable 5.2 as the advisor needs it. Type
claude --versionin your terminal to check, andclaude updateif you are behind. - Fable access on your plan. On some plans Fable usage is billed to usage credits, and Claude Code asks you to agree to that once before it will use Fable.
Step by step
Step 1: Open Claude Code in your project
Open your terminal, go to your project folder, and type claude. You can also open the project in the Claude desktop app.
Step 2: See which model you are on
Type /model and press Enter. A list of models appears, with the one you are using now highlighted. Press Escape to close it without changing anything.
Step 3: Pick the right model for the job
Type one of these, depending on what you are about to do:
/model haikufor a quick question or a one-line edit./model sonnetfor normal work: fixes, edits, most of your day. Start here if you are not sure./model opusfor a new feature or anything that has to be right the first time.
You can switch at any point in a session. The next message uses the new model.
Step 4: Turn on Fable as your advisor
Type /advisor fable. Claude Code replies with a line that starts with Advisor set to, followed by the model name. Now your main model does the work, and asks Fable for a plan when it reaches a hard decision.
If it tells you to run /model fable instead: that is the one-time usage-credits agreement. Type /model fable, choose to continue on Fable, then switch back with /model sonnet and type /advisor fable again.
Your choice is saved, so it stays on in future sessions until you turn it off.
Step 5: Work as usual
You do not have to do anything different. Claude decides when to ask Fable, usually before it commits to an approach, when an error keeps coming back, or before it says a task is done. When it does, you will see an Advising line in the chat. Press Ctrl+O to read what Fable said.
If you want a second opinion at a specific moment, just ask: “consult the advisor before you continue”.
Step 6: Turn it off for small jobs
Type /advisor off before a quick fix or a short task. This is the step most people skip, and it is the one that saves the most. The next section shows why.
Step 7: Check what you used
Type /usage at the end of a session. Advisor usage is counted in there too, so you can see whether it was worth it.
What I tested
I gave every model the same small coding task: write a scheduling function, write tests for it, and keep going until every test passes. One run each, in Claude Code 2.1.325, on November 18, 2026. Every model finished with all tests passing.
| Setup | Usage vs Sonnet | Time |
|---|---|---|
| Haiku 5.5 alone | Under a tenth | 45 seconds |
| Sonnet 5.5 alone | 1x | 27 seconds |
| Opus 5.5 alone | About 2x | 40 seconds |
| Fable 5.2 alone | About 4x | 49 seconds |
| Sonnet 5.5 with the Fable 5.2 advisor on | About 8x | 144 seconds |
Same right answer from all four models. For a job this size, the bigger models bought nothing extra.
The mistake I made
I left the advisor on for that same small task and asked Claude to check with Fable before starting and before finishing. It used about eight times what Sonnet used alone, and about double what Fable used by itself.
Here is why. Every time the advisor is called, Fable reads your whole conversation again, and that read is not cached. Two calls meant Fable read everything twice. Anthropic’s own documentation says the same thing: the advisor pays off on long, multi-step jobs where the plan decides the outcome, and adds less on short tasks where there is little to plan.
So my rule is simple: easy work on Sonnet, real builds on Opus, and the advisor on only when I am stuck or the job is long.
Common questions
Does it work in the Claude desktop app? Yes. The /advisor command works in the terminal, the desktop app and Remote Control.
Can Fable advise any model? Fable can advise Haiku, Sonnet and Opus. The advisor has to be at least as capable as your main model, so a smaller model cannot advise a bigger one.
How do I see which advisor is set? Type /advisor with nothing after it.
Does it work everywhere? It works on the Anthropic API and Claude subscriptions. It is not available on Amazon Bedrock, Google Cloud’s Agent Platform or Microsoft Foundry. It is also marked experimental, so details may change.
Limits of my test
This was one run per setup on one small task, so treat the numbers as a direction, not a benchmark. I told Claude to consult the advisor twice; left alone it may consult less. I have not tested a long refactor yet, which is where the advisor is meant to win. Run /usage on your own work before you trust my numbers or anyone else’s.
Keep reading
How to connect Claude to your WordPress site with WPVibe (2 minutes, free)
The exact steps I used to connect Claude and Claude Code to my WordPress site with the free WPVibe plugin. No passwords, no API keys, about two minutes.
Claude & AI 1 min readStop burning your Claude limit on Opus doing chores
Your Claude limit probably is not too small. Opus is doing work Haiku could do for a fraction of the cost, and the switch is already built into Claude Code.