Set the Reasoning Effort

Reasoning effort decides how much a model thinks before it answers. Kiro exposes five levels, and matching the level to the task is one of the easiest ways to control both speed and credit spend.

When to use this

Lower the effort for mechanical work: renaming things, writing boilerplate, formatting, small edits with a clear right answer. Raise it for design decisions, tricky debugging, and anything where a wrong answer costs more than the extra credits. Leave it alone when you are unsure; the default is reasonable.

Steps

1. Learn the ladder. The levels are low, medium, high, xhigh, and max. Lower means faster, shorter, and fewer credits.
2. Check your model supports the level you want. Claude Opus 5, Opus 4.8, Opus 4.7, Sonnet 5.5, and Sonnet 5 support all five. Opus 4.6 and Sonnet 4.6 have no xhigh. Opus 5.5 is not yet in Kiro’s effort table, so check what the panel offers. The GPT-5.6 models (Sol, Terra, Luna) support every level and default to high.
3. Change it in the IDE by clicking the model name in the chat input. That opens the panel where you pick the effort. In Kiro Web (since September 22, 2026), the effort selector sits next to the model and appears only for models that support it.
4. Change it in the CLI with /model (v2.23.0), which lets you switch models and set effort in one panel. /effort still works independently in V3 sessions and accepts a level directly (/effort low). Start a session with the --effort flag when you already know what the task needs.
5. Set a default so you stop deciding every time. Put it under chat.modelDefaults in ~/.kiro/settings/cli.json for a personal default, or .kiro/settings/cli.json for a project default.
6. Understand which setting wins. The order is session, then workspace, then user, then Kiro’s built-in default. A /effort change in a session overrides everything below it.
7. Watch the credit line. Higher effort means more thinking tokens, and thinking tokens are billed work like anything else.

Check it worked

Ask the same modest question at low and at high and compare. Low should come back faster and shorter. If you see no difference at all, check the model actually supports the level you set, and check nothing at a closer scope is overriding it. A session-level /effort silently wins over the file default.

Common problems

• The level did not apply. Your model does not offer it. xhigh is missing on Opus 4.6 and Sonnet 4.6.
• The default is ignored. Something closer in scope is set. Session beats workspace, workspace beats user.
• Credits climb. A high default across every session adds up. Set the default low and raise it per task.
• Answers got worse after you lowered it. That is the trade. Lower effort is for work where the answer is obvious, not for work you find hard.
• You edited the wrong file. ~/.kiro/settings/cli.json is your personal file. .kiro/settings/cli.json is the project one, and it wins for that workspace.

Related

• Change the default model is the other half of the same decision.
• Choosing a Model compares context windows and credit multipliers.
• Credits & Pricing explains how billed work is counted.