Models and Defaults

Choose model defaults that balance quality, speed, and cost

Overview

Model choices are practical trade-offs across quality, speed, and cost.

What to tune

  • Harness
  • Model
  • Effort level
  • Mode

Default strategy

  • Pick one default harness/model combo for most tasks.
  • Use higher effort only for hard reasoning or complex refactors.
  • Keep routine tasks on lower-latency settings.

Per-task overrides

Use overrides when a specific task has different needs:

  • higher reasoning depth for complex bug hunts
  • faster responses for simple repetitive edits
  • model swap when one provider is unstable

Change management

When changing team defaults:

  1. test on a few representative tasks
  2. compare quality, latency, and cost
  3. document why defaults changed