Change the Default Model

Kiro lets you pick which model answers, and the default is Auto. Changing it takes one click in the IDE, and the main decision is whether you have a reason to leave Auto at all.

When to use this

Change the model when a task has a shape Auto is not built for: an extremely long context, or bulk mechanical work where a cheap model is enough. Leave it on Auto the rest of the time. Auto carries a 1.0x credit multiplier, which is the baseline every other multiplier is measured against.

Steps

1. Open the chat input in the Kiro IDE and click the model name. The panel that opens is where you pick the model and the reasoning effort.
2. Choose a model, or leave Auto selected. Auto routes to frontier models with automatic fallback, uses intent detection and caching, and never routes to experimental models. On the Free plan it routes to Sonnet-class models or better, and on paid plans to Opus-class or better.
3. Check what the choice costs. Each model has a credit multiplier relative to Auto’s 1.0x, and a context window. Choosing a Model has the current table.
4. In the CLI, set your preference in a session. Kiro CLI has kept persistent model preferences since v2.6.0, so the choice carries into later sessions rather than resetting each time.
5. Set the reasoning effort separately. Effort is not the model, and lowering it is often the cheaper change. Use /effort in the CLI, or the same IDE panel. See Set the reasoning effort.
6. Store effort defaults under chat.modelDefaults in ~/.kiro/settings/cli.json for yourself, or .kiro/settings/cli.json for a project. Session settings beat workspace, workspace beats user, and user beats Kiro’s built-in default.

Check it worked

Send a short prompt and look at the model name shown in the chat input. It should match what you picked. Then start a brand new session and check again: if it reverted, you changed the model for one conversation rather than setting a preference. Watch your credit usage for a day after switching to a higher multiplier, because the difference is easy to miss until the bill arrives.

Common problems

• The model is not in the list. Model availability changes and some models are marked experimental. The current lineup lives at kiro.dev/docs.
• Credits drain faster than before. You moved from Auto’s 1.0x to a higher multiplier. Opus-class models sit at 2.2x and GPT-5.6 Sol at 2.4x.
• The answer got worse on a cheap model. Cheap models suit mechanical work. Move back to Auto for anything that needs judgement.
• The choice resets. You set it for one session rather than as a preference.
• You wanted faster answers, not a different model. Lower the reasoning effort instead. It is usually the better lever.

Related

• Choosing a Model compares every model, context window, and multiplier.
• Set the reasoning effort covers the other half of the decision.
• Credits & Pricing explains how multipliers turn into spend.