Stop paying Opus prices for Haiku work.

rightsize is a free Claude Code mod. It rates each prompt, matches the effort and model to the work, and shows you what it changed. You can undo any change.

Copy install promptView on GitHub

Work in progress

rightsize is a work in progress. It changes how Claude Code spends your tokens, and its choices are estimates. Savings are not guaranteed: some turns can cost more, not less, and a cheaper model can sometimes do weaker work. Use it at your own risk, and check /rightsize report to see what it really did for you. This is an independent project, not made or endorsed by Anthropic.

Give it a few turns to learn. rightsize starts with general estimates. It needs some turns of your real work before it knows how your prompts, your codebase, and your models behave. Don't expect the first few recommendations to save money, and some may cost a little more. As it compares each estimate with what you actually spent, its calls get better. Judge it after a few sessions, not a few prompts.

Found a bug, a wrong call or a saving that didn't show up? Open an issue on GitHub, or email rightsize@aivalue.blog.

Ask Claude to install it

Paste this into any Claude Code session.

Install the rightsize plugin from GitHub for me. Run these two commands in the terminal:
1. claude plugin marketplace add archid21/rightsize
2. claude plugin install rightsize@archid21 --scope user
Tell me if either one fails and why. When both succeed, tell me to start a new session and type /rightsize to check that it works.

Claude asks your permission before it runs each command. When it is done, start a new session and type /rightsize.

How it works

  1. It sizes each prompt. Before each prompt goes out, a quick, cheap Haiku check rates how hard the work is. Simple work runs at lower effort. Hard work gets more.
  2. It catches a stuck Claude. If the same command keeps failing, or Claude goes in circles, rightsize turns the effort up for the rest of that turn.
  3. It hands work to cheaper helpers. When a smaller model can do a task well, rightsize sends it to a helper (a subagent) on that model. Your main chat writes the brief, checks the result and asks for fixes. If the helper still can't get it right, the main chat does the work itself. It also moves Claude's own subagents to a cheaper model when that model can do the job.
  4. It paces your budget. On a subscription, it watches your 5-hour and weekly limits. If you use them up too fast, it saves harder. If you will have plenty left over, it lets Claude work harder.
  5. It learns and shows its work. After each prompt, it compares what it expected to spend with what you really spent, and adjusts. /rightsize report shows what it changed, about how much it saved, and what went wrong.

You stay in charge

Install in about 2 minutes

You need Claude Code 2.1.287 or newer in the terminal, or 2.1.286 or newer in the desktop app. Check with claude --version. If it is older, run claude update.

In the terminal

claude plugin marketplace add archid21/rightsize
claude plugin install rightsize@archid21 --scope user

Inside Claude Code

/plugin marketplace add archid21/rightsize
/plugin install rightsize@archid21

When it asks where to install, pick the user scope.

Then start a new Claude Code session. Sessions that were already open don't load it. Type /rightsize and press Enter. If it shows its status, it works.

Step-by-step guide in the README

Settings and privacy

Read the full Privacy section

FAQ

What does the rating call cost?

Each rating is one call to Haiku, the cheapest model. rightsize counts these calls: /rightsize report shows the saving after their cost. Prompts under 6 words are sent without a rating (you can change this in /config). Every 50th rated prompt is also checked by Sonnet in the background, and that cost is counted too.

Does it work on a subscription?

Yes. On a subscription, rightsize reads your 5-hour and 7-day windows and paces itself against them. Reports show a share of your usage, with dollar amounts marked as API-equivalent. On an API key, reports show dollars.

Does it send my code anywhere?

To rate each prompt, rightsize asks Claude Haiku through Claude Code, so your prompt goes to Anthropic, just as it does when you use Claude Code.

Nothing goes to rightsize's author unless you turn on anonymous feedback. Then it sends only numbers, such as task type, model and cost, to help make rightsize better for everyone. Never your prompts, code or files. Feedback is off by default.

Can I turn it off or undo a change?

Yes. Press Undo on the band, or type /rightsize undo, to run the last lowered prompt at your normal level. Each feature has a row in /config that you can turn off. /rightsize reset returns your settings to their defaults.

How do I turn it off or remove it?

Type /mods to switch it off or on at any time. To keep your settings but stop it loading, run claude plugin disable rightsize@archid21. To remove it, run claude plugin uninstall rightsize@archid21. The README lists what stays on your disk and how to delete it.

Which models does it know?

Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. These are built in, and the README explains how they are updated when models change. It works from the model and effort level your session starts on. It never changes your model itself.

Will it always save money?

No. Savings are estimates, and some turns can cost more. Check /rightsize report for what it did in your own sessions.