Models
Update

Effort levels: Light, Medium, High and Extra on every task

TL;DR

Effort levels let you choose how hard the agent works on a task: the composer's Effort chip offers Light, Medium (the default, unchanged), High and Extra. Effort never changes approvals, write limits or credit checks.

What each level changes

  • Light is cheaper and faster: less reasoning, fewer re-plans and repairs, shorter reports.
  • Medium is the default and behaves exactly as tasks did before.
  • High reasons more, keeps the plan critic always on, allows more re-plans and repairs, writes longer reports and lets an autonomous run take more steps.
  • Extra goes further on each of those, and may plan on a stronger model where the workspace kept the default models.

What it never changes

Approvals, write limits and credit checks are the same at every level. Effort is also available on schedules and through the API.

Frequently asked questions

Does higher effort cost more credits?
Usually, yes: more reasoning, re-plans and longer reports use more model tokens. Light is the cheaper, faster option; Medium is the default and unchanged.
Does Extra effort skip approvals?
No. Effort never changes approvals, write limits or credit checks.
Try it on your own account

Start free — 2,500 credits a month, no credit card. Reads are free, and every write waits for your approval.

Related releases