Young Tries AI

Cut Claude Code costs by letting open-effort pick the thinking level for each task

This guide is for developers who use Claude Code and want to lower costs. You install the open-effort plugin, watch it set effort per request, and learn how to keep it updated.

Saving moneyFor developers

Before you start

  • Claude Code installed. The one-line shell install needs the `claude` command in your shell; with only the desktop app, use the desktop-app route in step 3.
  • A project of your own to try the plugin in.
  1. There are three ways to install. The shell route is a single line and needs `claude` available in your shell. Run it in your terminal.

    Run in the terminal
    claude plugin marketplace add JakkapatB/open-effort && claude plugin install open-effort@open-effort
  2. If you are already in Claude Code in the terminal, type this command instead of using the shell route.

    Type in Claude Code
    /plugin install open-effort --marketplace JakkapatB/open-effort
  3. Go to Settings → Customize → Plugins. Click + Add → Add marketplace, type JakkapatB/open-effort and press Sync. In the Discover tab find Open effort, then choose Add to this computer ▾ → Install for me so it works in every project.

  4. Nothing else needs to be set up. Start a new session and look above the prompt.

    It worked if: You see a band above the prompt that shows the effort and model in use.

  5. In a project of your own, send a small request and watch the band. Haiku reads each request afresh, so a reading can differ a little from the examples. A clear rename should start at `low`, a task with a known goal at `medium`, and a hard bug at `xhigh`.

    Paste into Claude
    Rename getUser to fetchUser everywhere
  6. Now give Claude something unusually hard and compare the band with the one from the easy request.

    Paste into Claude
    Two orders can both take the last item in stock. Find the race and fix it
  7. When a test run fails, the effort card shows Check failed 1/3, then 2/3. On the third failure in a row the effort goes up one level and the card says where it came from (for example ↑ from low). The next passing run brings it back. If effort is already at the highest level, Claude asks whether to move up one model (Sonnet → Opus → Fable 5.1) and only moves if you say yes.

  8. open-effort checks for a new version once a day. When one is out, the band shows a highlighted ↑ Update button that opens the pane, with an Update now button at the top; after updating, the band's button says ↻ Reload. You can also update by hand in the shell, then start a new session.

    Run in the terminal
    claude plugin marketplace update open-effort && claude plugin update open-effort@open-effort
  9. In Claude Code in the terminal, run /plugin, open Marketplaces, pick open-effort and choose Enable auto-update. Or add the entry shown here inside ~/.claude/settings.json, next to any extraKnownMarketplaces entries already there.

    settings.json entry
    "extraKnownMarketplaces": { "open-effort": { "source": { "source": "github", "repo": "JakkapatB/open-effort" }, "autoUpdate": true } }

Watch out for

  • With only the desktop app installed, your shell has no `claude` command, so the one-line shell install will not work. Use the desktop-app route.
  • In the desktop app, typing /plugin install … only searches Anthropic's plugin directory, and open-effort is not listed there.
  • Claude Code auto-updates plugins only from Anthropic's own marketplaces, and the desktop app turns auto-update off for every marketplace. Use open-effort's own update button or the manual command.
  • The measured savings come from a small sample: a sample shop project, 61 sessions on Opus 5.5 at `high`. Each task ran once or twice, so cost figures vary from run to run.
  • Raising effort when stuck is not yet measured. Every test task passed on its first try, with and without the mod.

Questions

How much does it save?
In the author's test, 5 coding tasks run twice cost $5.65 with the mod off and $5.00 on (−11%), and all tasks still passed the hidden checks. An Explore sub-agent cost $0.41 on Opus at high versus $0.09 on Haiku at low (−78%). Cost is worked out from tokens at list price.
Will it switch to a stronger model on its own?
No. If the work needs a stronger model than the session's, Claude asks you first and only moves up with your yes. If checks still fail at the top, Claude stops and asks you what to do.
How does it choose the starting effort?
Haiku reads each new request. A clear plan starts at `low`, and a hard one starts at `xhigh`.
Does it support a larger workflow?
Yes. It can run Matt Pocock's Main Flow (grill → spec → tickets → implement → review → retro), with each stage and ticket at its own effort. Claude first asks whether to take a big task through it, with Matt Pocock's skills installed.

Sources

Based on